increase scikit-learn compatibility

[MarcBresson]

Hello,

I was working with CE transformers and stumbled upon an error in my workflow. After investigating, I found that CE transformers in ColumnTransformer object from scikit-learn do not necessarily transform all the given columns because of function category_encoders.utils.get_categorical_cols. Because of this inference of categorical columns, my integers column and boolean columns are not transformed.

To conform better with scikit-learn transformers, I think there should be a parameter that will allow the transformer to transform every columns. If we build on the existing cols parameter, we could have a special value cols="All" that triggers this behaviour.