Distributed Multi-Agent Reinforcement Learning with One-hop Neighbors and Compute Straggler Mitigation