ENH: Add annotations for np.core.shape_base

[BvB93]

This PR adds annotations for the following functions in np.core.shape_base:

• atleast_1d
• atleast_2d
• atleast_3d
• vstack
• hstack
• stack
• block

[rgommers]

Thanks @BvB93, LGTM.

[rgommers]

One thing that has become clear to me is that these kinds of type annotations are not useful as documentation. The type annotations attempt to be very correct for all sorts of corner cases, like axis: SupportsIndex, which makes them very confusing to the average user (who should just see int). If IDE's are going to use type annotations instead of docstrings for this, that's going to be problematic. I'm not sure we have discussed this issue elsewhere already?

[BvB93]

If IDE's are going to use type annotations instead of docstrings for this, that's going to be problematic.

I feel that a fundamental problem here is that many (seemingly simple?) runtime behaviors cannot be conveniently expressed with static typing, thus forcing us to resort to the likes of axis: SupportsIndex (or worse: overloads).

[rgommers]

I feel that a fundamental problem here is that many (seemingly simple?) runtime behaviors cannot be conveniently expressed with static typing

Yeah, this is partly historical. I'd much prefer to have axis only accept an int, and if static typing had existed when NumPy was first written, that's probably what the design would have been. Same with the array_like mess. Retrofitting static typing onto a too-forgiving code base is not ideal.

Maybe at this point it just is what it is, and it's on IDE's to do the right thing here (prioritize docs over annotations).