Learning Generalist Models for Anomaly Detection