Making it `scikit-learn` compatible

[jvdd]

I think that a great enhancement to this library would be to make it scikit-learn compatible. Basically this boils down to inherit from the SelectorMixin & BaseEstimator class & implement the .fit() and .transform() methods.

Once this library is scikit-learn compatible, you can include it in a Pipeline - resulting in fewer data leakage issues :)

An illustration of how this can be realized: https://github.com/predict-idlab/powershap/blob/main/powershap/powershap.py#L17