Prevent numpy error for block diag on Windows when numpy>=2.0

[jmamzax]

From version 2.0 NumPy on Windows uses int64 as default integer: https://numpy.org/devdocs/numpy_2_0_migration_guide.html#windows-default-integer. Checking NumPy version will prevent from the error AssertionError: Object comparison failed: torch.int64 != torch.int32 when NumPy will be updated on CI. Issue: #4326

[BBBela]

Looks good to me, thanks!

Functionally it is okay, but I thought about separating this conditional check - something like that:

        # numpy < 2.0 on Windows used int32 as the default integer type
        # numpy >= 2.0 uses int64 everywhere
        scipy_int_type = (
            torch.int32
            if IS_WINDOWS and np.lib.NumpyVersion(np.__version__) < "2.0.0"
            else torch.int64
        )

        expected_scipy_types = [
            torch.float64,
            scipy_int_type,
            torch.complex128,
            torch.float64,
        ]

But that is just a suggestion - approving anyway 😉