Train and test reinforcement learning agents.
Recent revisions are kept. Restoring one creates a new revision.