Personalizing Reinforcement Learning from Human Feedback with Variational Preference Learning