To Go Far, Go Together: Diverse Preferences Induce a Curriculum for Reward Optimization