Add squeeze(f, A, dims) for reductions to drop dims

[mbauman]

This simple definition makes it easier to write reductions that drop the dimensions over which they reduce. Fixes #16606, addresses part of the root issue in #22000.

[ararslan]

Bump. Good to go?

[mbauman]

Sure. It seems like a reasonable and wanted feature.

[mbauman]

It's been a minute. Let's re-run CI.

[ararslan]

Just realized it should probably have a NEWS item.

[mbauman]

NEWS added. This is good to go once CI passes.

[JeffBezanson]

I won't block this, but it's a little odd how this definition assumes the passed function uses the A, dims argument pattern. In general you might have to write

squeeze((a,d)->reducedim(max,a,d), A, dims)

[nalimilan]

Maybe it should always pass the function to reducedim?

[mbauman]

That's a really good point. Even worse is that it would prohibit us from transitioning sum(A, region) to a keyword argument for region within 1.x. Without it, we could deprecate the non-keyword method, but this function would always have to call the non-keyword method.

[lstagner]

What is the status of this? squeeze is now dropdims and I would very much like this feature.

[mbauman]

Closing in favor of #33130.