TypedPython: zero-copy buffer-protocol arrays (array.array, numpy, memoryview) as ArrayParam #149

Approved by the user (2026-10-04). Lists are copied in and written back (O(n) per call); numba and Cython read buffers in place. Accept exact array.array / numpy ndarray (contiguous, matching dtype) / memoryview parameters as native arrays without copy. Identity holds because no user code runs while the array is live (the existing ArrayParam contract), so direct writes are observed exactly when CPython would make them visible.

Done when: differential tests against the interpreter for each buffer type, non-contiguous and wrong-dtype buffers take the object path, a released/resized buffer cannot be used (PyBUF lifetime), and a call-heavy benchmark shows the copy gone.