ROCKET-1: Mastering Open-World Interaction with Visual-Temporal Context Prompting