mini_md (pybind11 interop demo)

Notes

• Pure compute loops release the GIL using py::gil_scoped_release.
• NumPy arrays are passed as contiguous float64 buffers; shapes are validated.
• The LJ kernel here is intentionally naive O(N^2) for clarity. Swap with neighbor lists for scale.