BLD: Support free-threaded CPython and ship 3.14t/3.15/3.15t wheels

[mbUSC]

• Declare the extension safe to run without the GIL with py::mod_gil_not_used (requires pybind11>=2.13)
• Replace psutil, which has no cp315t wheels, with os.process_cpu_count (3.13+) / os.cpu_count; n_threads=-1 now uses the logical cores available to the process
• Build cp314t, cp315 and cp315t wheels; drop cp38 wheels as cibuildwheel 4 no longer supports it (the sdist still supports 3.8)
• Bump cibuildwheel from 3.4.1 to 4.3.1 for stable CPython 3.15 support
• Add 3.14, 3.14t, 3.15 and 3.15t to the Linux and macOS test matrices
• Fix sdist include path (src/diptest_core -> src/diptest-core)

[mbUSC]

Hi @RUrlus 👋
I made this PR to ship the missing wheels for new Python versions, including free-threaded wheels.
Free-threaded execution is trivially safe here since the extension is made of pure functions with no shared mutable state (every call allocates its own buffers and RNG), and the Python layer only holds constants set at import. The tests pass on free-threaded Python 3.15.0 with the GIL confirmed off, and a 16-thread stress test matches the single-threaded results.

Please lmk if anything needs to be changed or if it can be merged -- thanks!

[RUrlus]

Hi @mbUSC,

Appreciate the PR and looks good. The PyBind11 bump is exposing a Valgrind build issue, I think this commit should take care of it. If so, I'll merge the PR tomorrow

[RUrlus]

It does, so I'll merge it in and create a release this week.
Thanks again