[rasidi3112]

Summary

GitHub's hosted runners have no GPU, so the real-CuPy tests are always skipped in CI. This adds notebooks/gpu_validation.ipynb, which can be opened in Google Colab on a free GPU runtime. The notebook:

1. runs the full test suite, including test_real_cupy_matches_cpu and the legacy tests/test_3d_gpu_solver.py;
2. benchmarks CPU vs GPU on 3D rotating Gross–Pitaevskii runs (64³–160³) and reports the largest |ψ_GPU − ψ_CPU|;
3. prints a summary for the README.

It also explains itself and skips the GPU parts when no NVIDIA driver is present, and it is safe to run more than once. The README GPU section gains an "Open in Colab" badge.

Test plan

• Notebook validates with nbformat
• Executed end to end locally without a GPU: GPU parts skipped with a clear message, 118 tests pass, CPU benchmark runs
• "Run all" executed twice in the same kernel without errors
• Run on a Colab GPU runtime (results to be added to the README)