A GAN-based data poisoning framework against anomaly detection in vertical federated learning