Understanding Generalization of Federated Learning: the Trade-off between Model Stability and Optimization