Build a Tool-Calling Agent Loop

Implement run_agent(question, model, tools, max_turns=5):

1. Send the question to the model, offering tools.specs().
2. Append the reply to the conversation.
3. If the model didn't ask for a tool, return response.text.
4. Otherwise run each requested tool and append one user message holding a tool_result block per call, in request order.
5. An unknown tool is not a crash. Return the error as the tool result with "is_error": True and let the model recover.
6. Stop at max_turns and return None.