REGR: Revert GH51335

[rhshadrach]

Ref: #51955 (comment)

• closes #xxxx (Replace xxxx with the GitHub issue number)
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

@jbrockmendel - this only reverts the change to _reduce, but leaves other changes untouched. Would you prefer to revert all of #51335?

[jbrockmendel]

Does the whatsnew from #51335 need to be changed/reverted?

[rhshadrach]

@jbrockmendel - yes, it did - thanks.

[jbrockmendel]

OK this looks fine. Let's reiterate the plan for when i inevitably forget: merge this, backport to 2.0, then for 2.1 do either #51923, which might need to be updated to un-revert some or all of this?

[rhshadrach]

I think you meant #51955 instead of #51923. #51955 can either go into 2.0 or 2.1 depending on whether there is time.

[phofl]

Is this ready for 2.0?

[rhshadrach]

I believe this is ready, wanted to make sure @jbrockmendel is on board with the plan. I think both this and #51955 would be good for 2.0.1.

[rhshadrach]

Ack - need to update the whatsnew, so no, this isn't ready anymore.

[jbrockmendel]

Yes I am on board with the plan.

[rhshadrach]

I'm a bit more negative on this now. With 2.0.0 now released, this with #51955 will now reintroduce bad behavior (test_minmax_tzaware_skipna_axis_1 is xfailed). I'm going to see if I can get that test working in #51955, and if this is the case then I think we can go foward.

Without being able to fix that test, I think this path should be abandoned and we should instead pursue (a) increase perf of transpose as @jorisvandenbossche has suggested (#52083) and (b) use online ops where possible as @jbrockmendel has suggested (#52083 (comment)).

[jbrockmendel]

I like this plan; I'm excited to be a part of it.

[rhshadrach]

I don't see any reasonable way to support axis=1 with tzaware dtypes by operating on NumPy arrays - so I think there is no way forward in this direction.