CI: add Python 3.15

[dnicolodi]

No description provided.

[rgommers]

I'll capture the test error on Windows with 3.15.0b2 for posterity:

 FAILED tests/test_sdist.py::test_symlinks - AssertionError: assert {'symlinks-1.0.0/meson.build', 'symlinks-1.0.0/submodule/bbb.py', 'symlinks-1.0.0/pyproject.toml', 'symlinks-1.0.0/__init__.py', 'symlinks-1.0.0/PKG-INFO', 'symlinks-1.0.0/submodule/aaa.py'} == {'symlinks-1.0.0/meson.build', 'symlinks-1.0.0/pyproject.toml', 'symlinks-1.0.0/submodule/bbb.py', 'symlinks-1.0.0/__init__.py', 'symlinks-1.0.0/PKG-INFO', 'symlinks-1.0.0/submodule/aaa.py', 'symlinks-1.0.0/submodule/__init__.py'}
  
  Extra items in the right set:
  'symlinks-1.0.0/submodule/__init__.py'
  
  Full diff:
    {
        'symlinks-1.0.0/PKG-INFO',
        'symlinks-1.0.0/__init__.py',
        'symlinks-1.0.0/meson.build',
        'symlinks-1.0.0/pyproject.toml',
  -     'symlinks-1.0.0/submodule/__init__.py',
        'symlinks-1.0.0/submodule/aaa.py',
        'symlinks-1.0.0/submodule/bbb.py',
    }

Fingers crossed the one-liner fix in python/cpython#151671 lands in beta 3.

[ngoldbaum]

Fingers crossed the one-liner fix in python/cpython#151671 lands in beta 3.

No such luck unfortunately. You'll need to wait for beta4, beta3 just came out: https://discuss.python.org/t/python-3-15-0-beta-3-is-here/107866

[dnicolodi]

The issue with symlinks in tar archives on Windows is mostly theoretical: I don't think there is anyone using symlinks on Windows withing source distribution of Python packages. We can just mark the test as a known failure an move on.

[rgommers]

I'd hope everyone uses Linux for publishing sdists indeed, but it's not completely theoretical, e.g. if Matplotlib (which uses symlinks) would run wheel builds as python -m build (i.e., via sdist), that'd go wrong on Windows.

Fine with marking as xfail or skip on win-315 and move on for now, but let's keep track of it to ensure we undo that before 3.15.0rc1

[dnicolodi]

Fine with marking as xfail or skip on win-315 and move on for now, but let's keep track of it to ensure we undo that before 3.15.0rc1

This implies that the bug will be fixed. I don't know the dynamics of the CPython development but, given the interested demonstrated so far, I doubt that it will be fixed.

[rgommers]

I have some more confidence in that, but worst case we'll close the tracker issue again. For now let's get this in once CI is green.