$\rm E(3)$-Equivariant Actor-Critic Methods for Cooperative Multi-Agent Reinforcement Learning