How to store large information received via API in the state in LangGraph, use it inside a tool, but not pass unnecessary data into the LLM context? For example, the LLM only sees IDs and names of specialists, while the tool gets their schedule and additional data.