In light of #6697

1. Should we have a BaseEstimatorCV class defining the required parameters (like results_, best_estimators_) as an interface attribute that can be overwritten?
2. Should the parameter still be called results_ and not search_results_ - Let's call it cv_results_ would be more representative of what it holds.

TODO

• calibration.py:class CalibratedClassifierCV(BaseEstimator, ClassifierMixin)
• covariance/graph_lasso_.py:class GraphLassoCV(GraphLasso):
• feature_selection/rfe.py:class RFECV(RFE, MetaEstimatorMixin):
• linear_model/coordinate_descent.py:class LassoCV(LinearModelCV, RegressorMixin):
• linear_model/coordinate_descent.py:class ElasticNetCV(LinearModelCV, RegressorMixin):
• linear_model/coordinate_descent.py:class MultiTaskElasticNetCV(LinearModelCV, RegressorMixin):
• linear_model/coordinate_descent.py:class MultiTaskLassoCV(LinearModelCV, RegressorMixin):
• linear_model/least_angle.py:class LarsCV(Lars):
• linear_model/least_angle.py:class LassoLarsCV(LarsCV):
• linear_model/logistic.py:class LogisticRegressionCV(LogisticRegression, BaseEstimator,
• linear_model/omp.py:class OrthogonalMatchingPursuitCV(LinearModel, RegressorMixin):
• linear_model/ridge.py:class RidgeCV(_BaseRidgeCV, RegressorMixin):
• linear_model/ridge.py:class RidgeClassifierCV(LinearClassifierMixin, _BaseRidgeCV):

@amueller @jnothman @MechCoder @vene @GaelVaroquaux @agramfort