Multi-Agent Manipulation via Locomotion using Hierarchical Sim2Real