TLeague: A Framework for Competitive Self-Play based Distributed Multi-Agent Reinforcement Learning