CMIX: Deep Multi-agent Reinforcement Learning with Peak and Average Constraints