Numpy 2.0 Compatibility

[jclds139]

A couple sets of changes needed for Numpy 2.0, just replacing np.unicode_ with np.str_ and np.NaN with np.nan. Both are drop-in replacements for Python >= 3.6, which looks like the new version requirement. Particularly since there seems to be a plan to drop Python 2 support, these shouldn't cause further issues.

I will note, two tests are still failing, but seem unrelated to these changes:

• test_has_required_lazy() in test_marshallers_reuiring_modules.py is throwing a tarfile-related error that looks like something related to pytest running tests asynchronously.
• test_numpy_sized_dtype_nested_0 in test_write_readback.py() is failing for all for format cases:
  • With no metadata or Python-only metadata, it's a ValueError that size must be positive. May be something from h5py
  • With MATLAB metadata (with or without Python metadata as well), it's a TypeNotMatlabCompatibleError for a void. I'm assuming that means an empty dataset was coerced to void type somewhere, but I haven't been able to find it yet.

[felixdittrich92]

Hi @frejanordsiek any updates here ? :smiley: