MoG-VLM: Efficient Video Anomaly Detection through Motion-Guided Vision-Language Models