Private Reinforcement Learning with PAC and Regret Guarantees