Optimism and Delays in Episodic Reinforcement Learning