Refactor insurance churn risk to use wave-ml

[vopani]

Using Wave ML

Bare minimum refactoring to use wave-ml for modelling. There is more that can be done but this is a good place to start. Part of #41

Have tested this locally ✅

cc: @mtanco @geomodular

[vopani]

Looks great Rohan. I see self.wave_model.predict() could have been used but test_dataset_path would have needed to be loaded twice because of predict_contribution(). We should be able to accept H2O frame, no?

The test dataset is also required in some other methods of the class like _get_explanation so instead of loading twice, I kept it separate.

I think we can do an updated refactor of this once we bring in the interpretability part into Wave ML, then we can remove all those other methods altogether and just use Wave ML for everything.

[vopani]

@mtanco This is ready to be merged. Things to note:

• Wave ML uses H2O-3 AutoML for modelling. Currently runtime set to 30 seconds - You can change this here

• Doesn't use validation split nor does cross-validation (didn't find it necessary) but I could add it if required.

• Models enabled are DRF and GBM. Enabling other models (like XGBoost or DeepLearning) breaks the current implementation of shapley, so please avoid.