IDEA: Saving data, Simple surrogates

Based on ML Coffee from 2021-11-19

We already collect actor data and objective data for SingleOptimizable. We can save these as numpy arrays. It's easy and cheap to train a Random Forest on this data and quickly get a reasonably accurate surrogate for SingleOptimizable environments.

The next step would be to also allow data collection and surrogate training for Gym environments. This might be a bit more difficult.