/v1/audio/speech returns 400 (not 404) when model is omitted

[thinmintdev]

What to build

POSTing /v1/audio/speech with a body that omits the model field returns HTTP 404 "File Not Found". OpenAI's reference returns 400 with a clear "you must provide a model parameter" message. The 404 looks like the route is missing and wastes debugging time.

Map the dispatcher's "no model in body" branch to 400 with an OpenAI-shaped message.

Acceptance criteria

• POST /v1/audio/speech without model returns 400, not 404
• Response body is OpenAI-shaped with a clear message naming the missing field
• The same fix is applied to any other /v1/* dispatch path that exhibits the same behaviour
• Unit test covers the model-missing case for /v1/audio/speech and the equivalent path on /v1/audio/transcriptions

Blocked by

None - can start immediately

---

Sourced from tests/harness/FINDINGS.md §18 (low).