Optimal Algorithms for Stochastic Contextual Preference Bandits