自主上下文压缩

A new tool has been added to the Deep Agents SDK (Python) and CLI, enabling models to compress their own context windows at opportune moments. Context compression is an action that reduces the amount of information in an agent's working memory, where older messages are replaced by newer ones to optimize information processing efficiency. This feature aims to enhance the flexibility and responsiveness of agents when handling complex tasks, ensuring they maintain efficient decision-making capabilities even in high-information scenarios. By autonomously managing context, models can better adapt to dynamic environments and improve overall performance.