Optimistic Reinforcement Learning with Quantile Objectives