Reinforcement learning support

[rohan843]

We can build some streaming training-inference system that can allow for applying reinforcement learning to a model from within the system.