Add/Fix Caching to ModelStacker

[shane-kercheval]

Cross Validation (and probably ModelFitter/etc.) have a build_cache_key function which gets the name of the model via model_name = type(model).__name__. This won’t work on ModelStacker, because there might be multiple ModelStackers, with different Base Models (and hyper params) and different stacking models (and hyper-params)

• [x] perhaps ModelWrapperBase has a property like model_name that does type(self).__name__ (not sure that will work; or something), and ModelStacker can override this.
• [x] test caching with ModelFitter
• [x] test caching with Resampler
• [x] test caching with ModelSearcher

final structure:

ModelStacker_[stacking model type]_[hyper_params].pkl  # final stacked model
base_[Base Model Description 1].pkl  # first base model (trained on all data)
base_[Base Model Description 1].pkl  # second base model (trained on all data)
...
/resample_[Base Model Description 1]/...  # files associated with resample models of first base model
/resample_[Base Model Description 2]/...  # files associated with resample models of second base model

[shane-kercheval]

also refactored ModelSearcher file caching structure to have a folder for each tuned model, and a file for each final model.