Driver Anomaly Quantification for Intelligent Vehicles: A Contrastive Learning Approach With Representation Clustering