COMBO: Compositional World Models for Embodied Multi-Agent Cooperation