MAINT: linalg: remove a stray np.cast

[ev-br]

Reference issue

numpy/numpy#24144 (comment)

What does this implement/fix?

Remove a stray call to np.cast.

Additional information

Now, without looking it up: what is the result of _I = np.cast['F'](1j) ?

[rgommers]

Thanks Evgeni.

Now, without looking it up: what is the result of _I = np.cast['F'](1j) ?

Got it wrong;) Honestly, I even had to look up dtype='F'. Always have to think hard about those letters, this should be spelled dtype=np.complex64. Also, the _I variable is only used in two places it looks like, so ideally it shouldn't have been there at all. But never mind, if needed that can be tweaked at a later date. This PR fixes the immediate issue.

[rgommers]

I also manually triggered the nightly wheel builds on main, to get a new nightly up quickly (may help JAX or scikit-learn).

[larsoner]

Thanks @rgommers we just hit this in mne-tools/mne-python#11760 (comment) !

[pllim]

Hello and thanks for this patch! Any chance scipy dev wheel can be updated with this included? Otherwise, I get import error if I test downstream with both numpy dev and scipy dev wheels.

[ev-br]

so ideally it shouldn't have been there at all.

Yes, could have been just 1j, but then combined with things being arrays or 0D arrays or numpy scalars or python scalars... Safer just keep it as it was :-).

[rgommers]

New wheels uploaded to https://anaconda.org/scientific-python-nightly-wheels/scipy/files minutes ago (some are still trickling in). Please let me know if that didn't do the trick.