Seeking Wisdom on Cuda

Hi! I am attempting to get this working with Cuda support. Any ideas? Thank you! Works great without the CUDA flag.

**Hardware:**
- RTX480 with latest drivers
- CUDA 13.0 toolkit
- Windows
- Cudarc v17.7 (If this is a limfac, I can run a local patch etc to sync these)
- CUDA kernels known working from non-ML parts of the application with Cudarc and cuFFT bindings.

**Cargo.Toml**:
```toml

candle-core = { git = "https://github.com/huggingface/candle", package = "candle-core", features = ["cuda"] }
candle-nn = { git = "https://github.com/huggingface/candle", package = "candle-nn", features = ["cuda"] }
```

**The compile error:**
```rust
error: failed to run custom build command for `candle-kernels v0.9.2-alpha.1 (https://github.com/huggingface/candle#d4545ebb)`
note: To improve backtraces for build dependencies, set the CARGO_PROFILE_RELEASE_BUILD_OVERRIDE_DEBUG=true environment variable to enable debug information generation.
Caused by:
process didn't exit successfully: `C:\Users\the_a\code\Bio\daedalus\target\release\build\candle-kernels-b4b265fd7ce556ce\build-script-build` (exit code: 101)
--- stdout
cargo::rerun-if-changed=build.rs
cargo::rerun-if-changed=src/compatibility.cuh
cargo::rerun-if-changed=src/cuda_utils.cuh
cargo::rerun-if-changed=src/binary_op_macros.cuh
cargo:info=["/usr", "/usr/local/cuda", "/opt/cuda", "/usr/lib/cuda", "C:/Program Files/NVIDIA GPU Computing Toolkit", "C:/CUDA"]
cargo:rerun-if-env-changed=CUDA_COMPUTE_CAP
cargo:rustc-env=CUDA_COMPUTE_CAP=89
cargo::warning=Builder { cuda_root: None, kernel_paths: ["src\\affine.cu", "src\\binary.cu", "src\\cast.cu", "src\\conv.cu", "src\\fill.cu", "src\\indexing.cu", "src\\quantized.cu", "src\\reduce.cu", "src\\sort.cu", "src\\ternary.cu", "src\\unary.cu"], watch: [], include_paths: ["src\\binary_op_macros.cuh", "src\\compatibility.cuh", "src\\cuda_utils.cuh"], compute_cap: Some(89), out_dir: "C:\\Users\\the_a\\code\\Bio\\daedalus\\target\\release\\build\\candle-kernels-e32097457b6dd4eb\\out", extra_args: [] }
--- stderr
thread 'main' panicked at C:\Users\the_a\.cargo\registry\src\index.crates.io-1949cf8c6b5b557f\bindgen_cuda-0.1.5\src\lib.rs:306:40:
Could not find CUDA in standard locations, set it manually using Builder().set_cuda_root(...)
stack backtrace:
0: std::panicking::begin_panic_handler
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\std\src\panicking.rs:697
1: core::panicking::panic_fmt
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\core\src\panicking.rs:75
2: core::panicking::panic_display
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\core\src\panicking.rs:268
3: core::option::expect_failed
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\core\src\option.rs:2139
4: core::option::Option::expect
5: bindgen_cuda::Builder::build_ptx
6: build_script_build::main
7: std::rt::lang_start
8: std::rt::lang_start
9: std::rt::lang_start
10: std::rt::lang_start_internal::closure$0
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\std\src\rt.rs:175
11: std::panicking::catch_unwind::do_call
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\std\src\panicking.rs:589
12: std::panicking::catch_unwind
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\std\src\panicking.rs:552
13: std::panic::catch_unwind
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\std\src\panic.rs:359
14: std::rt::lang_start_internal
at /rustc/1159e78c4747b02ef996e55082b704c09b970588/library\std\src\rt.rs:171
15: std::rt::lang_start
16: main
17: invoke_main
at D:\a\_work\1\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:78
18: __scrt_common_main_seh
at D:\a\_work\1\s\src\vctools\crt\vcstartup\src\startup\exe_common.inl:288
19: BaseThreadInitThunk
20: RtlUserThreadStart
note: Some details are omitted, run with `RUST_BACKTRACE=full` for a verbose backtrace.
```

I believe this may be due to a limitation of the `bindgen_cuda` dependency, which doesn't look in the correct location for CUDA, and doesn't check the PATH (Or LD_LIBRARY_PATH on Linux) environment variables. I have been compiling CUDA using `nvcc` in build.rs, and it is able to find CUDA without specifying any paths.