Scalable Multi-Agent Reinforcement Learning through Intelligent Information Aggregation