Fix deprecation of shape assignment in numpy 2.5+

[mscheltienne]

x-ref: mne-tools/mne-python#13585

Fixes:

/opt/hostedtoolcache/Python/3.14.2/x64/lib/python3.14/site-packages/nitime/utils.py:759: in tapered_spectra
    t_spectra.shape = rest_of_dims + (K, NFFT)
    ^^^^^^^^^^^^^^^
E   DeprecationWarning: Setting the shape on a NumPy array has been deprecated in NumPy 2.5.
E   As an alternative, you can create a new view using np.reshape (with copy=False if needed).

And other occurrences.

[mscheltienne]

Needs the same cmopatibility function as MNE for numpy < 2.1.

[mscheltienne]

This should be better, it adds however packaging as a dependency which should be harmless and provides better version comparison than manual string parsing.

[larsoner]

Looks good to me but I'll let @effigies merge if he's happy!

[effigies]

In #205 we settled on using SPEC 0, which says we should be dropping numpy<2 now anyway. I think we can get rid of all the compat stuff and just use the numpy 2 features. The minimum nibabel following SPEC 0 rules would be 5.3 (see https://github.com/nipreps/speczeroplus/blob/main/spec-0000/chart.md).

[mscheltienne]

However, this requires numpy 2.1.

[larsoner]

@mscheltienne I think @effigies is suggesting removing some of the existing shims (some of which were slightly tweaked here anyway) while you're messing around with this code

[mscheltienne]

Happy to remove the other elements which are in for numpy < 2 and to then bump the pin to 2.0+.

[effigies]

As to reshape, let's just use a = a.reshape(newshape). The default behavior is to create a view if at all possible, which it should be in all of these cases.

[mscheltienne]

@effigies Done! numpy 2.0+, compatibiltiy code gone, and simpler a = a.reshape(newshape) with fallback of copy=None (default).