A Survey of In-Context Reinforcement Learning