DEPR: groupby.idxmin/idxmax will all NA values

[rhshadrach]

#54234 deprecated the case of skipna=False and any NA value, but for consistency with Series/DataFrame versions, we also want the case of skipna=True with all NA values in a group to also raise. E.g. this should raise a ValueError

df = pd.DataFrame({'a': [1, 1, 2], 'b': np.nan})
gb = df.groupby("a")
print(gb.idxmin())

[jbrockmendel]

Make this a breaking change in 3.0? Could do a full deprecation cycle, but the inconsistency will be annoying in the interim.

[rhshadrach]

That sounds good to me.