Refactor extra.numpy.basic_indices() handling of max_dims when allow_newaxis=False

[Zac-HD]

Via #3065 (comment), basic_indices() is doing more work than it needs to when allow_newaxis=False - and that's particularly salient because the array_api extra doesn't have newaxis at all. So the internals part:

• When allow_newaxis=False, max_dims should be set to len(shape), and
• We can therefore move it to _array_helpers.py and refactor the array_api extra to use this too

This also suggests a constraint on the arguments that we're not currently enforcing for the Numpy extra:

if max_dims > len(shape) and not allow_newaxis:
    note_deprecation(..., since="RELEASEDAY", has_codemod=False)
    # Becomes `raise InvalidArgument(...)` in the next major version

[honno]

Happy to do this—will make a PR next week.

For reference this was the deprecation notice I had written before. I think it covers the right things, I just need to iterate.

>>> nps.basic_indices((3, 3, 3), max_dims=5).example()
HypothesisDeprecationWarning: max_dims=5 and allow_newaxis=False, so we cannot generate indices that would produce arrays with dimensions greater than len(shape)=3 i.e. the dimensions of the indexed array. In the next release, max_dims would default to the the dimensions of shape=(3, 3, 3) when allow_newaxis=False, and passing max_dims greater than len(shape) will raise an error.

[Zac-HD]

How about f"max_dims={max_dims} is larger than len(shape)={len(shape)}, but allow_newaxis=False makes it impossible for an indexing operation to add dimensions."? Trying to mention all the relevant arguments and their relationship early, and then provide a short explanation afterwards.

[Zac-HD]

Closed by #3098.