[BUG FIX] Fix GPU detection in test infrastructure for WSL2

[Lidang-Jiang]

Summary

• Fix _torch_get_gpu_idx() crashing with FileNotFoundError on WSL2 due to missing /proc/driver/nvidia/gpus/
• Wrap /proc/driver/nvidia/gpus/ access in try/except, falling back to existing defaults
• Remove unused variable nvidia_gpu_indices in _get_gpu_indices()

Root cause

WSL2 provides full CUDA support (nvidia-smi works, torch.cuda.is_available() returns True), but the native Linux NVIDIA driver sysfs interface at /proc/driver/nvidia/gpus/ does not exist. Two functions in tests/conftest.py rely on this path:

1. _get_gpu_indices() — guarded by os.path.exists(), silently falls back to (0,) (correct for single GPU)
2. _torch_get_gpu_idx() — calls os.listdir() without existence check, causing a crash

Fix

Wrap both /proc/driver/nvidia/gpus/ accesses in try/except FileNotFoundError, falling back to the already existing defaults ((0,) and -1). Native Linux behavior is unchanged.

Verification

Tested on WSL2 (Windows 11, NVIDIA GeForce RTX 3050 Ti Laptop GPU, Driver 581.83, CUDA 13.0).

Before (original code on WSL2)

============================================================
BEFORE: Original _get_gpu_indices() and _torch_get_gpu_idx()
============================================================
Platform: linux
/proc/driver/nvidia/gpus/ exists: False

torch.cuda.is_available(): True
torch.cuda.get_device_properties(0).name: NVIDIA GeForce RTX 3050 Ti Laptop GPU

_get_gpu_indices() = (0,)  (fell through to default (0,), did NOT detect via /proc)

_torch_get_gpu_idx(0) = CRASH! FileNotFoundError: [Errno 2] No such file or directory: '/proc/driver/nvidia/gpus/'

After (fixed code on WSL2)

============================================================
AFTER: Fixed _get_gpu_indices() and _torch_get_gpu_idx()
============================================================
Platform: linux
/proc/driver/nvidia/gpus/ exists: False

torch.cuda.is_available(): True
torch.cuda.get_device_properties(0).name: NVIDIA GeForce RTX 3050 Ti Laptop GPU

_get_gpu_indices() = (0,)  (try/except caught FileNotFoundError, fell through to default)

_torch_get_gpu_idx(0) = -1  (try/except caught FileNotFoundError, fell through to default)

pytest --backend gpu on WSL2 (after fix)

============================= test session starts ==============================
platform linux -- Python 3.11.14, pytest-9.0.2, pluggy-1.6.0 -- python3
rootdir: worktree-genesis-wsl2
configfile: pyproject.toml
plugins: xdist-3.8.0, anyio-4.13.0, syrupy-5.1.0, forked-1.6.0
collecting ... collected 1 item

tests/test_sensor_camera.py::test_destroy_unbuilt_scene_with_camera
[Genesis] [INFO] Running on [NVIDIA GeForce RTX 3050 Ti Laptop GPU] with backend gs.cuda. Device memory: 4.00 GB.
[Genesis] [INFO] Genesis initialized. version: 0.4.3, precision: 32
PASSED

============================== 1 passed in 7.87s ===============================

Test plan

• pytest --backend gpu passes on WSL2 (previously crashed)
• pytest --backend cpu passes (regression test)
• Native Linux: /proc/driver/nvidia/gpus/ path still used when available (no behavior change)

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[duburcqa]

Try catch and fallback to the already existing default is sufficient. Do not use nvidia-smi.

[Lidang-Jiang]

You're right, much simpler. Removed all nvidia-smi fallback and switched to try/except with existing defaults in 9a244a9.