DOC: ndimage: improve grayscale morphology docstrings

[grlee77]

The goal of this MR is to make the behavior of the structure argument clearer and avoid confusion as seen, for example, in #19384

[grlee77]

Looks like there was an unrelated failure in linalg:

scipy/sparse/linalg/tests/test_expm_multiply.py:35: in wrapped
    with pytest.warns(UserWarning,
E   RuntimeWarning: overflow encountered in multiply
        args       = (<5x5 MatrixLinearOperator with dtype=float64>, array([[ 1.22796439, -1.14500358],
       [-1.02018788, -0.39673356],
       [ 0.71869298, -0.84965727],
       [-0.36083862, -0.3083584 ],
       [-0.84761337, -1.80172566]]))
        func       = <function expm_multiply at 0xe588bc40>
        kwds       = {'endpoint': True, 'num': 2, 'start': 0.1, 'stop': 3.2}
=========================== short test summary info ============================
FAILED scipy/sparse/linalg/tests/test_expm_multiply.py::TestExpmActionInterval::test_expm_multiply_interval_matrix
= 1 failed, 49340 passed, 2580 skipped, 11447 deselected, 566 xfailed, 12 xpassed in 632.01s (0:10:32) =

[lucascolley]

@RSKothari does this help explain things?

[rgommers]

LGTM, thanks @grlee77. This is a clear improvement, so let's get this in right before the release branch creation, so it ends up in docs users will see soon.

[RSKothari]

@lucascolley Yes it helps. Thank you for the improvements.