* fixed bug in choosing best model
* breaking change of MedValueNaive, moved into new AvgValueNaive
* increased speed by only saving detailed model results when ensembling
* improved handling of data with large numbers of NaN, prints warning now if NaN in any test set
* experimental option to save model results during run, allowing progress recovery if crashes
* added max_per_model_class option to prevent one model from predominating after many iterations
* added weekly sample data (from EIA)
* added GluonTS models
* added TSFreshRegressor (albeit in a very slow version)