Improve local developer offline run experience

[dtzar]

trafficstars

A developer should be able easily to run and debug the solution from their local development machine which shows up as an "offline run" in the Azure Machine Learning SDK, potentially against the actual Azure ML service run for debugging.

Today there is duplicated code supporting this experience in these locations:

1. verify_train_pipeline
2. evaluate_model
3. register_model

This code should be centralized so that all of the various python files support the OfflineRun experience. This should also be coordinated with #139