Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback