ComfyUI startup crash on ROCm torch 2.5.1+rocm6.2 with comfy_kitchen 0.2.28 (torch custom op schema infer_schema: kernel_size list[int])

### Custom Node Testing

- [x] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

Environment

GPU: RX 7600
ROCm/hip: 6.2.41133-dd7f95766 (rocm 6.2)
PyTorch: torch==2.5.1+rocm6.2
Python: 3.12
ComfyUI commit: fe4195f7f4275f2626cbafc703acc3ddde1e5490
comfy_kitchen: comfy_kitchen==0.2.28

Crash details
ComfyUI crashes during import/startup when comfy_kitchen registers a PyTorch custom op via `torch.library.custom_op

Top-level error:
ValueError: infer_schema(func): Parameter kernel_size has unsupported type list[int]

Failing function signature:
`(q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, kernel_size: list[int], is_causal: list[bool], scale: float | None) -> torch.Tensor)`

Failing decorator location (from traceback):
/comfy_kitchen/backends/eager/na.py on:
@torch.library.custom_op("comfy_kitchen::na3d", mutates_args=())

Full traceback (as reported)
text

Traceback (most recent call last):
File "/mnt/Docker/ComfyUI/main.py", line 237, in
import comfy.utils
File "/mnt/Docker/ComfyUI/comfy/utils.py", line 25, in
import comfy.memory_management
File "/mnt/Docker/ComfyUI/comfy/memory_management.py", line 8, in
from comfy.quant_ops import QuantizedTensor
File "/mnt/Docker/ComfyUI/comfy/quant_ops.py", line 23, in
import comfy_kitchen as ck
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/comfy_kitchen/__init__.py", line 3, in
from .backends import cuda as _cuda_backend # noqa: F401
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/comfy_kitchen/backends/cuda/__init__.py", line 161, in
from comfy_kitchen.backends.eager import rope as _eager_rope # noqa: E402
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/comfy_kitchen/backends/eager/__init__.py", line 71, in
from .na import na3d
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/comfy_kitchen/backends/eager/na.py", line 163, in
@torch.library.custom_op("comfy_kitchen::na3d", mutates_args=())
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/torch/_library/custom_ops.py", line 133, in inner
schema_str = torch.library.infer_schema(fn, mutates_args=mutates_args)
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/torch/_library/infer_schema.py", line 106, in infer_schema
error_fn(
File "/mnt/Docker/comfyenv/lib/python3.12/site-packages/torch/_library/infer_schema.py", line 58, in error_fn
raise ValueError(
ValueError: infer_schema(func): Parameter kernel_size has unsupported type list[int]. The valid types are: dict_keys([...]). Got func with signature (q: torch.Tensor, k: torch.Tensor, v: torch.Tensor, kernel_size: list[int], is_causal: list[bool], scale: float | None) -> torch.Tensor).

Reproduction steps

Install PyTorch 2.5.1+rocm6.2
Install comfy_kitchen==0.2.28
Start ComfyUI at commit fe4195f7f4275f2626cbafc703acc3ddde1e5490
ComfyUI crashes immediately at import due to torch.library.custom_op schema inference rejection of `kernel_size: list[int]

Expected behavior

ComfyUI should start normally with comfy_kitchen 0.2.28 on ROCm PyTorch 2.5.1.

### Actual Behavior

Actual behavior

ComfyUI fails at startup due to PyTorch custom-op schema inference error for comfy_kitchen::na3d.

### Steps to Reproduce

Update ComfyUI through Comfy Manager

### Debug Logs

```powershell
The trace is posted above. Logs have been overwritten after reinstalling ComfyUI
```

### Other

Had to force 0.2.27 comfy_kitchen to get the app to run.

Start with comfy/quant_ops.py, which imports comfy_kitchen, then inspect the installed comfy_kitchen backends/eager/na.py decorator shown in the traceback. Reproduce the import crash with PyTorch 2.5.1+rocm6.2 and comfy_kitchen 0.2.28, comparing behavior with 0.2.27. Done means ComfyUI starts normally on the reported environment.

Written by the indexing model from the issue text.