MAINT, TST: two types of failures observed on maintenance/1.13.x branch

[tylerjereddy]

They seem pretty clearly unrelated to the diff in gh-20567, but one of them was fixed with quite a large PR on main and not sure we'd want to backport that.

• MacOS: https://github.com/scipy/scipy/actions/runs/8813658434/job/24191865773?pr=20567 which is basically RuntimeWarning: invalid value encountered in _beta_ppf that was fixed by MAINT/BLD: Remove stats._boost and add the distribution related functions to scipy.special._ufuncs #20393 -- do we really want/need to backport the ~3k line diff? I didn't check the diff in detail yet, but special has been so active that merge conflicts may be likely if there are changes there.
• Windows: https://github.com/scipy/scipy/actions/runs/8813658421/job/24191867663?pr=20567 FAILED linalg/tests/test_decomp.py::TestOverwrite::test_eig_bandedz ; our issue tracker is empty for test_eig_banded; I do note that I see scipy.linalg._testutils._FakeMatrix2 in the traceback, and we did have to make some minor shims to machinery like that when dealing with the NumPy array copy semantics changes, but pure speculation that this would be the cause