test: add naive i64 matmul kernel

[tetsuo-cpp]

Summary

• Add a naive i64 matmul pipeline test (M=2, N=3, K=4)
• Add a GPU end-to-end matmul kernel test
• Document the one-thread-per-output work partition in the test

Test plan

• cmake --build build --target format
• uv run ruff check gpu_test/
• uv run ruff format gpu_test/
• cmake --build build --target check-warpforth
• GPU tests: uv run pytest -v -m gpu gpu_test/test_kernels.py