Enhancements

• #184 Change Keras+TF Neural Networks to scikit-learn MLP
• #233 Limit staking number of classes and models
• #232 Remove Linear model from Compete mode
• #208 Improve importance computation for large number of columns
• #205 Remove small learning rates for Xgboost

Bug fixes:

• #231 Restricted characters in feature_neams in Xgboost
• #227 Fix strings in golden_features.json - thank you @SuryaThiru!
• #215 Assure at least 20 samples (or k_folds) for each class

Docs update:

• #213 Update docs in AutoML - thank you @shahules786!