Multi-Agent Reinforcement Learning for Pragmatic Communication and Control