Automatically keep categorical type on merge

[Gerenuk]

Code Sample, a copy-pastable example if possible

import pandas as pd
dd1=pd.DataFrame({"a":pd.Series(["x"], dtype="category")})
dd2=pd.DataFrame({"a":pd.Series(["x", "y"], dtype="category")})
dd3=dd1.merge(dd2)
dd3.dtypes  # categorical turns into object type

Problem description

It would be nice if when merging on categorical columns, their categorical nature would remain by an automatic union_categoricals. Currently only categoricals with identical values remain categorical and merge on non-identical categoricals becomes object type.

[TomAugspurger]

We have a similar open issue for pd.concat to union the categoricals. No one has implemented it yet.

[ms7463]

I believe this would be relevant to this PR also. The feedback I got seemed to suggest that categorical with different dtypes should not be considered compatible for set operations.