Evolutionary Reinforcement Learning for Sample-Efficient Multiagent Cooperation