FakeTensor construction dispatches `aten::detach` into a Python PrivateUse1 backend's kernel

### 🐛 Describe the bug

For a backend registered through `_setup_privateuseone_for_python_backend`, `FakeTensorMode` cannot wrap the output of the first op it runs on one of the backend's tensors. We get
```
RuntimeError: Creating a new Tensor subclass FakeTensor but the raw Tensor object is already associated to a python object of type Tensor which is not a subclass of the requested type
```

Using `with FakeTensorMode() as mode:` is enough to trigger it.

This is also the source of a dynamo skip in the pytorch test suite: https://github.com/pytorch/pytorch/blob/ca66d9844119d868477faa88c782bd46495e1783/test/test_privateuseone_python_backend.py#L166-L167

### Reproducer

Uses the same `torch._C._acc` helpers as https://github.com/pytorch/pytorch/blob/ca66d9844119d868477faa88c782bd46495e1783/test/test_privateuseone_python_backend.py

```python
import torch
from torch._subclasses.fake_tensor import FakeTensorMode
from torch.utils.backend_registration import _setup_privateuseone_for_python_backend

_setup_privateuseone_for_python_backend("foo")

@torch.library.impl("aten::empty.memory_format", "privateuseone")
def empty(size, *, dtype=None, layout=None, device=None, pin_memory=None, memory_format=None):
print(f"called empty with device {str(device)}")
return torch._C._acc.create_empty_tensor(size, dtype or torch.get_default_dtype())

@torch.library.impl("aten::detach", "privateuseone")
def detach(self):
print(f"called detach with device {str(self.device)}")
if self.device.type == "foo":
return torch._C._acc.create_empty_tensor(self.shape, self.dtype)
# A meta tensor reached the backend's kernel. Do the reasonable thing anyway.
return torch.empty_like(self)

x = torch.empty((3, 4), device="foo")

with FakeTensorMode() as mode:
fx = mode.from_tensor(x)
fx.view(-1)
```

Output on torch 2.14.0:

```
called empty with device foo
called detach with device meta
Traceback (most recent call last):
File "/home/gabriel/projects/playground/main.py", line 27, in
fx.view(-1)
~~~~~~~^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_compile.py", line 54, in inner
return disable_fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_dynamo/eval_frame.py", line 1548, in _fn
return fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/utils/_stats.py", line 29, in wrapper
return fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 1707, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 2495, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 1855, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 3174, in _dispatch_impl
op_impl_out = op_impl(self, func, *args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_impls.py", line 267, in dispatch_to_op_implementations_dict
return op_implementations_dict[func](fake_mode, func, *args, **kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_impls.py", line 1240, in _view_meta
torch._refs._reshape_view_helper(a, *shape, allow_copy=allow_copy),
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_refs/__init__.py", line 4208, in _reshape_view_helper
return torch.as_strided(a, [a.numel()], [1])
~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/utils/_stats.py", line 29, in wrapper
return fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 1183, in __torch_dispatch__
return func(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_ops.py", line 916, in __call__
return self._op(*args, **kwargs)
~~~~~~~~^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_compile.py", line 54, in inner
return disable_fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_dynamo/eval_frame.py", line 1548, in _fn
return fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/utils/_stats.py", line 29, in wrapper
return fn(*args, **kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 1707, in __torch_dispatch__
return self.dispatch(func, types, args, kwargs)
~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 2495, in dispatch
return self._cached_dispatch_impl(func, types, args, kwargs)
~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 1855, in _cached_dispatch_impl
output = self._dispatch_impl(func, types, args, kwargs)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 3216, in _dispatch_impl
self.wrap_meta_outputs_with_default_device_logic(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
r, func, flat_args, device=kwargs.get("device")
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
)
^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 3363, in wrap_meta_outputs_with_default_device_logic
return tree_map(wrap, r)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/utils/_pytree.py", line 1584, in tree_map
return treespec.unflatten(map(func, *flat_args))
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/utils/_pytree.py", line 1313, in unflatten
leaves = list(leaves)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 3353, in wrap
out = converter.from_meta_and_device(
self, e, device or common_device
)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 698, in from_meta_and_device
out = FakeTensor(
fake_mode, t, device, pytype=pytype, dispatch_keys=dispatch_keys
)
File "/home/gabriel/projects/playground/.venv/lib/python3.14/site-packages/torch/_subclasses/fake_tensor.py", line 1026, in __new__
self = Tensor._make_subclass(
cls,
...<3 lines>...
device_for_backend_keys=device,
)
RuntimeError: Creating a new Tensor subclass FakeTensor but the raw Tensor object is already associated to a python object of type Tensor which is not a subclass of the requested type
```

## Proposed fix

I'm happy with either of those. The first is the more general one.

**In `_make_subclass`.** When `device_for_backend_keys` is given, perform the
detach under a `c10::impl::ExcludeDispatchKeyGuard` for that device's backend
key. The caller is asking for a raw `TensorImpl` with the backend keys *set on
the result*, and the existing stash-and-disable guards two lines above show that is the intent.
This fixes every subclass built this way.

**In `FakeTensor.__new__`.** Since `elem` is asserted to be on the meta device
a few lines later, exclude the backend key of `device` around the
`_make_subclass` call.

### Versions

Collecting environment information...
PyTorch version: 2.14.0+cu126
Is debug build: False
CUDA used to build PyTorch: 12.6
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.5 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04.3) 11.4.0
Clang version: Could not collect
CMake version: Could not collect
Libc version: glibc-2.35

Python version: 3.14.0 (main, Oct 10 2025, 12:47:49) [Clang 20.1.4 ] (64-bit runtime)
Python platform: Linux-5.15.0-1108-nvidia-x86_64-with-glibc2.35
Is CUDA available: False
CUDA runtime version: No CUDA
CUDA_MODULE_LOADING set to: N/A
GPU models and configuration: No CUDA
Nvidia driver version: No CUDA
cuDNN version: No CUDA
Is XPU available: False
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: False
Caching allocator config: N/A

CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Gold 6430
CPU family: 6
Model: 143
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 2
Stepping: 8
CPU max MHz: 3400.0000
CPU min MHz: 800.0000
BogoMIPS: 4200.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf tsc_known_freq pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 cat_l2 cdp_l3 invpcid_single cdp_l2 ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities ibpb_exit_to_user
Virtualization: VT-x
L1d cache: 3 MiB (64 instances)
L1i cache: 2 MiB (64 instances)
L2 cache: 128 MiB (64 instances)
L3 cache: 120 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42,44,46,48,50,52,54,56,58,60,62,64,66,68,70,72,74,76,78,80,82,84,86,88,90,92,94,96,98,100,102,104,106,108,110,112,114,116,118,120,122,124,126
NUMA node1 CPU(s): 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43,45,47,49,51,53,55,57,59,61,63,65,67,69,71,73,75,77,79,81,83,85,87,89,91,93,95,97,99,101,103,105,107,109,111,113,115,117,119,121,123,125,127
Vulnerability Gather data sampling: Not affected
Vulnerability Indirect target selection: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Reg file data sampling: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced / Automatic IBRS; IBPB conditional; PBRSB-eIBRS SW sequence; BHI BHI_DIS_S
Vulnerability Srbds: Not affected
Vulnerability Tsa: Not affected
Vulnerability Tsx async abort: Not affected
Vulnerability Vmscape: Mitigation; IBPB before exit to userspace

Versions of relevant libraries:
[pip3] Could not collect
[conda] Could not collect

cc @chauhang @penguinwu @eellison @aorenste @liangel-02 @bdhirsh @bobrenjc93 @NmomoN @mengpenghui @fwenguang @cdzhan @1274085042 @PHLens @albanD

Start with torch/_subclasses/fake_tensor.py, especially FakeTensor.__new__ and the from_meta_and_device path shown in the traceback. Review test/test_privateuseone_python_backend.py around lines 166-167 and reproduce the PrivateUse1 case; done means FakeTensorMode can wrap the backend tensor and run view(-1) without dispatching detach into the backend kernel or raising the subclass-construction error.

Written by the indexing model from the issue text.