ENH: ndimage: add axes support to most morphology functions

[grlee77]

Reference issue

A series of previous MRs (#18016, #18261, #18305, #20809, #20811) addes axes support to ndimage functions in _filters.py.

This MR extends the same axes support to most functions in _morphology.py. All grayscale and binary morphology functions are covered, only the three distance_transform_* functions are not updated here.

The changes needed to support axes were relatively small as all of these functions are based on either (_filters._min_or_max_filter or _binary_erosion). Given that _min_or_max_filter was already updated to support axes previously, _binary_erosion is the main function updated here.

cc: @ev-br, @lucascolley, @mdhaber and @dschmitz89 who had helped review some of the prior MRs.

What does this implement/fix?

The functions updated in this MR fall into two groups:

1. The grayscale morphology functions are based on calling _filters._min_or_max_filter which already has axes support, so those were trivial to update here. These include grey_dilation, grey_erosion, grey_opening, grey_closing, morphological_laplace, morphological_gradient, white_tophat and black_tophat.

2. The binary morphology functions are all based on _binary_erosion, which has axes support added to it in this MR. These include binary_erosion, binary_dilation, binary_opening, binary_closing, binary_hit_or_miss, binary_propagation and binary_fill_holes.

Additional information

The testing approach follows that taken in the other recent MRs adding axes support to the filtering functions.

[grlee77]

@lucascolley : The Array API case is happy now.

There was one seemingly unrelated failure in the "Oldest GCC" case, but I am rerunning that one now to see if it is reproducible

[ev-br]

assert_array_almost_equal in this test file (and elsewhere in ndimage on main) is a thin wrapper of xp_assert_close already: https://github.com/scipy/scipy/blob/main/scipy/ndimage/tests/test_morphology.py#L4

[ev-br]

Where does this PR stand @grlee77?

[ev-br]

Please, not this pattern again :-).
Now that skip_xp_backends decorators can be stacked, we can express things is a less confusing way. So these tests need to skip all non-cpu backends, and additionally you skip cupy and jax.numpy on CPU, is this correct?
Then,

@skip_xp_backends(cpu_only)
@skip_xp_backends("cupy", reason="these filters do not yet have axes support in CuPy")
@skip_xp_backends("jax.numpy", reason="these filters do not yet have axes support in JAX.numpy")

[grlee77]

Thanks, I will update it

[ev-br]

CI is green modulo an unrelated failure in lapack wrappers on MacOS / OpenBLAS, let's land this. Thank you @grlee77 , all.

[ev-br]

@grlee77 mind adding a release note snippet for the upcoming 1.15 release please?
https://github.com/scipy/scipy/wiki/*-Release-Note-Entries-for-SciPy-1.15.0-(asterisk-makes-this-appear-at-the-top)

[grlee77]

@grlee77 mind adding a release note snippet for the upcoming 1.15 release please?

Done. Thanks @ev-br, @lucascolley and @tylerjereddy for reviewing