Reward is enough: LLMs are in-context reinforcement learners