Distributional Preference Learning: Understanding and Accounting for Hidden Context in RLHF