Add optional GPU execution via CuPy

[01-1]

Summary

• add opt-in CuPy execution for selected backend NumPy calls while preserving CPU ndarray outputs and FLOP accounting
• expose configure_gpu, gpu_status, gpu_enabled, and gpu_available at the top level
• add a gpu optional dependency extra and README documentation for local GPU execution
• add focused tests for public controls, CPU array contract preservation, fake-CuPy offload, and CUDA runtime fallback

Testing

• UV_CACHE_DIR=/tmp/flopscope-uv-cache uv run --python 3.10 pytest -q tests/test_gpu.py