Store best models and internal model training on persistent store

[ardunn]

Persistent pipelines should store the best models (ie., top N) alongside the actual best one. The pipeline should also store the optimization history of the TPOT model