Add more regularization options to LogisticRegressionClassifier

[freddyaboulton]

Our LogisticRegressionClassifier only supports an l2 penalty, but the underlying sklearn implementation additionally supports l1, and elasticnet.

I think we should add these regularizer options to our component hyperparameters.

One possible complication is that certain regularizers can only be used with certain solvers.

[dsherry]

Sounds good. Let's add this to our component. If we don't get clear test results adding this to automl too, no need for now, can wait for new algo.