Long-Running Agentic Query Execution

Today there is an /ask endpoint backed by an ask service. The ask service queries an LLM API and a search API, and query execution happens in memory with no additional durability. Treat the LLM and search APIs as black boxes.