Introduce allow_newaxis to xps.indices()

[honno]

The Array API spec recently updated to allow expanding dimensions via None (data-apis/array-api#408), so this PR adds support for dim-expanding indexing in xps.indices() via an allow_newaxis kwarg, similiar to how nps.basic_indices() works.

Note numpy.array_api arrays don't support newaxis yet (numpy/numpy#21373), so test_generate_valid_indices[numpy] will fail when smoking the indexer. Fortunately the mock does support it, so HYPOTHESIS_TEST_ARRAY_API=all pytest tests/array_api/test_indices.py comes in quite handy here.

[honno]

Note numpy.array_api arrays don't support newaxis yet (numpy/numpy#21373)

I didn't realise CI was already testing numpy.array_api, neat! I've added a runtime xfail if xp == numpy.array_api... that's probably preferable to xp.__name__ == "numpy.array_api" I think? I put it in a try: ... except ImportError: pass block to work if you have older NumPy versions installed (or don't have any installed), which still seems like a good idea.

Assuming this gets patched, the next NumPy release following that patch we can check for numpy.__version__ first. Note most array_api patches seem to get backported. I guess we want to pin NumPy to the patched version rather than faffing around with this.

[Zac-HD]

Another fantastic PR @honno - thanks again!

One comment below about how we can ensure that our test is weakened as little as possible while np.array_api doesn't yet support newaxis (including fail-when-added, so that we undo the patch as soon as possible); otherwise this looks merge-ready to me. Congrats!

[honno]

Awesome @Zac-HD, thanks!