Locally Differentially Private Reinforcement Learning for Linear Mixture Markov Decision Processes