Pre-training as Batch Meta Reinforcement Learning with tiMe