A model emits three parallel tool calls: `get_inventory`, `get_price`, and `get_shipping`. To cut latency, the host appends each result to the transcript as it arrives and immediately re-invokes the model. Runs start failing with malformed-request errors, and some produce hallucinated shipping costs. What should the host do?

1. A Keep streaming each result back as it lands, tagging each with its call ID so the model can match results to requests on its own turn by turn.
2. B Execute the calls sequentially in emitted order so results arrive in a predictable sequence and the transcript stays consistent.
3. C Apply a short timeout and return only the results that finished in time, omitting slow calls so the turn is never blocked.
4. D Run them concurrently with per-call timeouts, then return all three results in one turn, keyed by call ID, with errors as explicit results.