make dims (previously sometimes called region) a keyword argument

[JeffBezanson]

Continues #25831, fixes #25501.

I still think the keyword argument is a better API, but after working on this I understand why @ararslan's opinion shifted. The basic issue is that reducing over dimensions generally requires a different algorithm and (in our case) gives a different type of result, so it's useful to dispatch on whether a dims argument is passed. Since we don't dispatch on keyword arguments, quite a few contortions are needed.

I found a total of 22 affected functions. Somebody who understands the stats functions better should review to make sure those changes make sense. There are also a few un-exported functions (covm, covzm, corm, corzm, reducedim!, mapreducedim!, maybe a few more) that take dim arguments, but I left those alone to save time. We might want to change them too though.

[ararslan]

I resolved the conflicts and the CI failures are unrelated. Good to go?

[JeffBezanson]

Thanks.

I'm still wondering about the un-exported stats functions. Do people use them enough that it's worth changing them to the new style?

[ararslan]

They're used in StatsBase but that's the only place I've seen them. Since they aren't exported I guess we would consider them "internal" in some sense, so it seems okay to me not to change them. Presumably if they were used that often they would be exported.

I'd also be interested to hear what @andreasnoack's thoughts on that are.