Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition