Momentum for the Win: Collaborative Federated Reinforcement Learning across Heterogeneous Environments