Learning Synthetic Environments and Reward Networks for Reinforcement Learning