Local Differential Privacy for Regret Minimization in Reinforcement Learning