WHL: add support for Python 3.15t+

[neutrinoceros]

try to leverage PEP 803
This might not work out of the box yet since NumPy needs to be built from source in CI.

[neutrinoceros]

Only one arch (macOS+arm64) is getting to the test phase and then fails miserably with

TypeError: 'ndarray' object is not an instance of 'ndarray'
while processing 'texture'

I think I'll merge the supporting refactor but will leave the enabling as a draft.

[neutrinoceros]

current status:

• all windows archs fail to build numpy 2.4.6
• 3 other archs (manylinux-arm64 + both macOS archs) fail tests in manner I described above

in any case I do not intend to merge this (even partially) before 3.15.0rc1 is available, and I expect that numpy might unblock everything once it has wheels for it on PyPI too.
If by then I still get failures, I'll do a partial merge with just the successful builds.

[neutrinoceros]

current status:

• cp315-cp315t nightlies are now available and correctly selected on Linux and MacOS, though the failure mode I reported previously is still showing, which probably indicates a bug in numpy-rust
• windows still doesn't have compatible nightlies, because of a known build time error upstream

Meanwhile, the "Future Python dependencies" job is failing to select a nightly wheel, supposedly because 2.5.0rc0 is prefered as it doesn't require forking on windows.

[neutrinoceros]

regarding forking, I don't think uv has a strategy that would enable what I'm looking for, so I'll just use the low level interface (uv pip) to force it.

edit: #442

ref: https://docs.astral.sh/uv/reference/settings/#fork-strategy

[neutrinoceros]

though the failure mode I reported previously is still showing, which probably indicates a bug in numpy-rust

given that #442 ran successfully with cp315 (GIL), I'd suspect pyo3 over numpy-rust, though it might be a combination of both, given how recent they are.

[neutrinoceros]

The good news is that I can easily reproduce the runtime error locally, so I should be able to produce a minimal report

[neutrinoceros]

PyO3/rust-numpy#555

[neutrinoceros]

This is now working as intented. The one thing I want to wait for before I undraft is a rust-numpy release with updated numpy headers, which this branch is currently using a dev branch for.