MAINT refactoring get_feature_names_out to define it in encoders

[rcap107]

Remove get_feature_names_out from the SingleColumnTransformer class and add it to the specialized encoders since they might specialized depending on their own parameters.

[rcap107]

This is the first part of an issue on refactoring the get_feature_names_out function. Here I moved it out of the SingleColumnTransformer, later we might want to create a class that abstracts it. We can't use the SingleColumnTransformer for that because not all the encoders that inherit from this use the same get_feature_names_out

[glemaitre]

As a first step, it looks good. We can further see if we want to limit the code redundancy but since the remaining duplicated code are in different files, I'm not sure that in terms of maintenance and code complexity, it is better to get an abstraction just for the get_feature_names_out and encoder exposing a n_components_ attribute.

[glemaitre]

Just pushed a small fixed to make sure that _SplineEncoder follows the scikit-learn API (not overwrite attributes) and make sure that we cover the line that was not covered before.