Patent Figure Classification using Large Vision-language Models