### PR summary

Closes #31909.

This PR makes numpy.lib._datasource accept pathlib.Path and other string path-like local paths in the same places that already accepted string paths. The affected datasource entry points now normalize path inputs with os.fspath() before the existing URL, cache, absolute-path, and file-opening logic runs.

Previously, direct pathlib.Path inputs could reach string-specific operations such as path.split(...) without conversion. This affected DataSource.abspath, DataSource.exists, DataSource.open, the module-level numpy.lib._datasource.open, and Repository path resolution. The change is intentionally scoped to datasource path normalization; URL handling, cache lookup order, compressed-file opener selection, write-mode URL rejection, and sandboxing logic are left unchanged.

The runtime tests cover Path inputs for DataSource.open, DataSource.exists, DataSource.abspath, Repository.abspath, Repository.exists, Repository.open, and the module-level open helper. The typing fixtures now use _typeshed.StrPath so pathlib.Path is accepted while bytes-path cases remain rejected.

Validation:

• Direct regression script using Python 3.12.13 passed for the affected Path cases and preserved URL-string detection (failure_count 0).
• python -m py_compile numpy/lib/_datasource.py passed with Python 3.12.13.
• git diff --check passed.
• spin build --clean was attempted locally, but this Windows environment hit MSVC fatal error C1001 internal compiler errors while compiling unrelated linalg sources, so I could not run the focused pytest target from a completed local build.

First time committer introduction

I am contributing a small bug fix to align DataSource behavior with its documented path-like support.

AI Disclosure

AI tools were used to help identify the failing path-like case, draft and validate the patch, and prepare this PR description. The proposed code changes and PR text were reviewed before submission.