Save agent progress

[AntoineRoll]

We have to find a solution to save agent's state during or after training to be able to use it after closing cloud notebook or restart training.

We need to clearly state the versions of agents (date, HP chosen, training parameters, etc).

Ideas:

• In TF, you can use a save_weights function to save a model's state.

[rarick]

Incrementally save model parameters to file for later use. Must be compatible with loading in #5

[rarick]

see model.save(filepath) in openAI baselines

[rarick]

This also needs to save the environment configuration which was used to produce the model. Save the yaml alongside these parameters.