Attention-based Fault-tolerant Approach for Multi-agent Reinforcement Learning Systems