Add numpy dtype access for TypeValue

[cjdoris]

Closes #737

Motivation

• Provide a simple Python-side mapping from primitive Julia TypeValue objects to the corresponding NumPy dtype to aid interoperability (buffers/array conversions).

Description

• Add a @property __numpy_dtype__ on TypeValue (in src/JlWrap/type.jl) that imports numpy and returns numpy.dtype(...) for the basic Bool, Int*, UInt*, Float*, Complex{Float*} (where supported) and Ptr{Cvoid} types.
• ComplexF16 (Complex{Float16}) is not mapped because NumPy does not support a complex32 alias here, so it is intentionally left unsupported.
• Unsupported/unknown Julia types raise AttributeError from __numpy_dtype__.
• Add pytest coverage in pytest/test_all.py verifying correct dtype mappings and that unsupported types (e.g. ComplexF16, String) raise AttributeError.

Testing

• Ran julia --project -e 'using Pkg; Pkg.test()', which completed successfully (registry update produced a warning but tests passed).
• Ran julia --project -e 'using Pkg; Pkg.test(coverage=true)', which completed successfully (registry update warning reported).
• Ran Python tests with uv run pytest -s --nbval ./pytest; an initial failure surfaced (NumPy does not accept a "complex32" alias), the implementation and tests were adjusted, and the final uv run pytest -s --nbval ./pytest completed with 19 passed.
• Ran coverage uv run pytest -s --nbval --cov=pysrc ./pytest, which completed successfully and reported the Python-side coverage summary (overall pysrc coverage reported by the run).

---

Codex Task