FEAT - Implement LogisticRegressionCV and ElasticNetCV?

[mathurinm]

For sparse penalties, Using warm start along the path gives quite a boost compared to using GridSearchCV

This feature was asked here : mathurinm/celer#302 (comment)

We can't have as many *CV classes as estimators, but ElasticNet and LogisticRegression seem popular enough to justify having those hardcoded (with both enet penalty).

Another solution would be to have GeneralizedLinearEstimatorCV, but it's not abvious at the moment how we would define the parameters to loop over (if the penalty is not always L1plusL2)

@wdyt @Badr-MOUFAD @QB3