Towards Real-Time Anomaly Detection within X-ray Security Imagery: Self-Supervised Adversarial Training Approach