Handle Tool Result Messages

To complete the loop, the API must be able to accept the result of a tool call from the user and feed it back to the model to generate the final human-readable response.

**Acceptance Criteria:**
- The `/v1/chat/completions` endpoint can now accept a `messages` list that includes a message with `role: "tool"`.
- This tool message (containing the result from the function the user executed) is correctly formatted and added to the conversation history sent to the model.
- The model then uses this result to generate a final text answer, and the endpoint returns a standard response with `finish_reason: "stop"`.

The work is in the `/v1/chat/completions` endpoint. Look at the request handling and message list validation. The model interaction logic needs to accept a `role: "tool"` message, format it correctly into the conversation history, and ensure the model uses it to generate a final answer with `finish_reason: "stop"`. Check existing tests for the endpoint to understand the expected request/response format.

Vom Indexierungsmodell aus dem Issue-Text verfasst.