[2306.14892] Supervised Pretraining Can Learn In-Context Reinforcement Learning