Add pickle support to estimators / pipeline

[rtanase]

Is your feature request related to a problem? Please describe.
Currently, only Nimbus Pipeline supports save_model / load_model, but if we just use a Nimbus learner within a SciKit pipeline then that resulting pipeline can't be successfully serialized/deserialized through pickle, because nothing is actually taking care of the the underlying binary file of the learner.

Describe the solution you'd like
I would like every Nimbus estimator & pipeline to properly deal with serialization/deserialization following pickle protocols. Under the hood, when pickle is invoked for an estimator, it could read its underlying binary model file and store it inside the state dictionary and during unpickling it could repopulate the binary file(through on override of __getstate__ and __setstate__ methods). Upon deserializing, the original binary file path might not be available, so it might be necessary to generate a new path for it and dump the contents there. The same can be implemented for Pipeline itself, not just for individual estimators.

Describe alternatives you've considered
Pickle

Additional context
N/A