MA2QL: A Minimalist Approach to Fully Decentralized Multi-Agent Reinforcement Learning