DM²: Decentralized Multi-Agent Reinforcement Learning for Distribution Matching