Differentially Private Reward Estimation with Preference Feedback