MetaDrive: Composing Diverse Scenarios for Generalizable Reinforcement Learning - View it on GitHub