Add linear regression with scikit learn

[eckelsjd]

Proposed Changes

• Use linear models from sklearn.linear_model, including Ridge, RidgeCV, Lasso etc. as an Interpolator alternative to Lagrange
• Optionally include polynomial features
• Optionally include sklearn preprocessors (e.g. MinMaxScaler) for normalization
• Includes new tests/docs for new functionality

Limitations

• Polynomial regression blows up in dimension for higher degree and larger numbers of inputs
• Linear regression doesn't play super nicely with SparseGrid training data -- i.e. you are learning a fit over all the inputs, while SparseGrid only samples over subsets of the inputs. Would recommend a new TrainingData implementation such as LHS.