PEFR/BUG/API: rename_categories with a series

[TomAugspurger]

In [2]: pd.Categorical(['a', 'b']).rename_categories(pd.Series([0, 1]))
Out[2]:
[a, b]
Categories (2, object): [a, b]

In 0.20.3 that was

[0, 1]
Categories (2, int64): [0, 1]
In [9]:

Do we want to treat series like an array or a dictionary here?

(this is causing the perf slowdown in rank I think).

[TomAugspurger]

xref #17586 implementing the dict-like rename.

[TomAugspurger]

I made #17982 assuming we want to use treat series as an array-like. I think that's the right move here, for backwards compat, but it's a bit unfortunate to have to special case it here.