explain_weights should support RFE/RFECV and other metaestimators

[hermidalc]

RFE/RFECV are not only feature selectors (SelectorMixin) but also classifiers/regressors (MetaEstimatorMixin), though ELI5 explain_weights doesn't support them as classifiers/regressors. The final fit of an RFE/RFECV object is a fitted estimator with either rfe.estimator_.coef_ or rfe.estimator_.feature_importances_ and in sklearn you do not usually follow up RFE/RFECV with another classifier or regressor. Would be great if ELI5 could support metaestimators in explain_weights.