Evaluating Query Efficiency and Accuracy of Transfer Learning-based Model Extraction Attack in Federated Learning