Learning Laparoscope Actions via Video Features for Proactive Robotic Field-of-View Control