Deception-Based Defense Against Model Poisoning Attacks in Federated Learning Using Generative Adversarial Network (GAN)