Add CUDA training setup documentation

docs/cuda_setup.md

@@ -0,0 +1,21 @@
+## CUDA Training Setup
+
+Luminal supports CUDA for accelerated ML training. To enable and debug CUDA workflows:
+1. Ensure you have the latest NVIDIA drivers and the CUDA toolkit installed (>= 11.x).
+2. Set the environment variable `CUDA_HOME` to your CUDA installation path.
+3. When using GitHub Actions, use a self-hosted runner with GPU. Example snippet:
+    ```
+    runs-on: self-hosted
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up CUDA
+        run: sudo apt-get install -y cuda-toolkit-11-2
+      - name: Test CUDA build
+        run: make test-cuda
+    ```
+4. For troubleshooting, run:
+    ```
+    nvidia-smi
+    python -c "import torch; print(torch.cuda.is_available())"
+    ```
+5. If you encounter build errors, see `docs/troubleshooting.md`.