Optimizing Communication and Device Clustering for Clustered Federated Learning with Differential Privacy