A Multi-agent Reinforcement Learning Approach for Efficient Client Selection in Federated Learning