Second Categorical fixups

[jankatins]

Docs:

• np.asarray(s2.cat) -> np.asarray(s2)
• Series(.....dtype='categorical') example, ENH: add support dtype='category' in Series constructor #8075
• Add docstring about internal useage to to_dense and from_array
• go through the docs and change levels -> categories
• clarify docs on usage of Categorical (public API restricted to creation (to specify levels/ordered, or to put in dataframe) don't use its methods/properties (-> s.cat is for that), apart from from_codes and maybe in a "under the hood" section )
• underline in categoriecal.rst

Code

• cut fixups, see API: update pd.cut with the new categorical integration #8077
• change __repr__ of Categorical (Discussion: feedback on the Categorical integration #8074 (comment))
• rename levels -> categories; add FutureWarning if levels is used in constructor and accessor
• reorder_levels and remove_unused_levels not inplace by default (and add inplace keyword, default False)
• s.unique() should return an array instead of index (-> np.asarray(self.levels))
• Add Series.cat.codes (wrapped into a Series)
• remove compat kwarg (+ update docs on this -> use from_codes)
• add rename/add/remove/reorder only methods and a "takes it all" method (probably set_levels(new_levels, rename=False)
• make assigning to categories strict (no additions/removals)
• fix for Problem in BUG: inconsistent behaviour of Groupby with empty bins when grouping by single/multiple keys #8138
• Failing Test on python 3.4

[jankatins]

I will look at these ToDos on a train ride tomorrow.

[jorisvandenbossche]

@JanSchulz I should also be able to find some time this week if you can use some help.

[jreback]

closed by #8153