Addressing Membership Inference Attack in Federated Learning with Model Compression