REGR: disallow mean of period column again

[jorisvandenbossche]

This reverts parts of #32426 and #29941 (those are only on master, not yet released), to disallow taking the mean on a Period dtype column again. The mean for period is not supported on PeriodArray itself or for a Series of period dtype (on purpose, see discussion in #24757 when mean for datetimelike was added), so for DataFrame columns it should also not work.

See also comment at #32426 (comment)

cc @jbrockmendel

[jbrockmendel]

Does this also disable median for PeriodDtype?

[jorisvandenbossche]

Does this also disable median for PeriodDtype?

Hmm, yes, indeed.

Now, I would say that can be left out of this PR, because:

• It's actually not even tested, and only enabled on master (no regression compared to released version)
• This PR disables it in nanops, but the follow-up PR REGR: fix DataFrame reduction with EA columns and numeric_only=True #33761 to call EA._reduce in frame._reduce will make median not work as well (even if it keeps working in nanops.nanmedian), since PeriodArray._reduce doesn't handle it.
• Currently the support for median is also inconsistent (xref BUG: inconsistent median support for datetimelike dtypes #33760), since it works for DataFrame when specifying numeric_only=False, but not for Series. Which again needs PeriodArray._reduce to implement median

So properly supporting median basically needs PeriodArray._reduce to support median, and while this is certainly something we should do, my feeling is that this is better done in a dedicated PR for that (with appropriate testing).

[jorisvandenbossche]

@jbrockmendel does that sound OK to you?
(and other comments on the code changes itself?)

[jbrockmendel]

does that sound OK to you?

Yes, this is a Worth Doing It Right situation.

[jreback]

lgtm. this does not test / disable median?

[jorisvandenbossche]

See my answer to Brock above

[jreback]

kk great @jorisvandenbossche