REF: Avoid special-casing in core.apply

[rhshadrach]

Ref #39935 (comment)

Currently in apply, there are many separate places where post-processing logic is applied. Once this is logic has been better consolidated, we can avoid special-casing certain ops (e.g. "size" for DataFrame).

[jbrockmendel]

Tangentially related: we've deprecated silent dropping of nuisance columns in groupby, resample, and DataFrame reductions, but apparently can also do so in core.apply. Would it make sense to deprecate it there too?

[rhshadrach]

Yes, I think so.

[rhshadrach]

@jbrockmendel I think we should also be doing this for transform. Here is an issue caused by fallback because of a typeerror:

df = pd.DataFrame(
    {'a' : ['x', 'y'], 'b' :['z', 'w'], 'c' : [1, 2], 'd' : [2., 5.]}
)
gb1 = df.groupby('a')
gb2 = gb1[['c', 'd']]
print(gb1.transform(lambda x: x.max() - x.min()))
print(gb2.transform(lambda x: x.max() - x.min()))

   c    d
0  0  0.0
1  0  0.0
     c    d
0  0.0  0.0
1  0.0  0.0