arxiv Revisiting Some Common Practices in Cooperative Multi-Agent Reinforcement Learning