Emergent Coordination and Phase Structure in Independent Multi-Agent Reinforcement Learning