CI: New NumPy release breaking Numba in our CI

[datapythonista]

pandas/tests/groupby/aggregate/test_numba.py:19: in <module>
    numba = pytest.importorskip("numba")
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E   pytest.PytestDeprecationWarning: 
E   Module 'numba' was found, but when imported by pytest it raised:
E       ImportError('Numba needs NumPy 2.2 or less. Got NumPy 2.3.')
E   In pytest 9.1 this warning will become an error by default.
E   You can fix the underlying problem, or alternatively overwrite this behavior and silence this warning by passing exc_type=ImportError explicitly.
E   See https://docs.pytest.org/en/stable/deprecations.html#pytest-importorskip-default-behavior-regarding-importerror

Source: https://github.com/pandas-dev/pandas/actions/runs/15541539524/job/43753280349#step:9:76

Somehow related: numba/numba#10105

@pandas-dev/pandas-core I guess it's not the first time this happens, since seems that Numba has been raising ImportError instead of pining the dependencies for at least one more version. Doesn't seem like we should be pinning NumPy ourselves. Any idea how this was fixed before if it already happened, or what should we do to fix the CI?

[TomAugspurger]

This was probably fixed by conda-forge/numba-feedstock#157.

I think conda-forge/conda-forge-repodata-patches-feedstock#1039 will fix this with no change needed on our part, or you could temporarily add that pin until that repo data fix is in.

[datapythonista]

Ah, amazing, thanks a lot for the info!