The Difficulty of Passive Learning in Deep Reinforcement Learning