.str._validate should infer for Series, not raise for all-na Index

[h-vetinari]

While working on #22725, a couple of work-arounds were necessary to correctly raise on wrong data types hiding as objects, e.g.

>>> pd.Series([1, 2, 3], dtype=object).str.cat([1, 2, 3])

However, already the .str accessor itself should raise on __init__ resp. the internal _validate method (this is closely related to #23011), i.e. instead of

>>> pd.Series([1,2,3], dtype=object).str
<pandas.core.strings.StringMethods object at 0x000002A4C70AE198>

it should be

>>> pd.Series([1,2,3], dtype=object).str
AttributeError

Interestingly, Index does correctly infer already in .str._validate:

>>> pd.Index([1,2,3], dtype=object).str
AttributeError: Can only use .str accessor with string values (i.e. inferred_type is 'string', 'unicode' or 'mixed')

However, there is another nit about that that I want to fix at the same time as the inferral for Series - namely that an all-na object Index (or Series) should not raise the AttributeError (currently it does because all-na gets inferred as float). There are legitimate cases where a selection of string data may be all-na (by alignment or whatever), and if the dtype is object then this shouldn't fail.

Edit: xref #9343 #13877