Theoretical considerations for practical meta reinforcement learning