Verify configs prior to training

[smithcommajoseph]

It is possible for an operator to create a model config that is invalid. At time of writing the team has identified time-based data_processing props that can be in conflict with each other, there may be other props as well.

We can bury the config verification step and call it in existing, exposed methods (prep_for_rnn, create_model, train, and the like), or we can expose the validation method separately for operators to call after configs are read into memory.