Classifier Clustering and Feature Alignment for Federated Learning under Distributed Concept Drift