Simultaneously Learning Stochastic and Adversarial Episodic MDPs with Known Transition