createTextToSpeech with stream:true answers SSE, but only the buffered JSON response is modelled — no CreateTextToSpeechAsStreamAsync

Problem

POST /v1/t2a_v2 (createTextToSpeech) accepts stream: true plus stream_options (TtsStreamOptions is in the spec), but src/libs/MiniMax/openapi.yaml declares only

responses:
  "200":
    content:
      application/json:
        schema:
          $ref: '#/components/schemas/TextToSpeechResponse'

With stream: true the API answers text/event-stream — one data: {…} event per audio chunk (data.audio hex, data.status), the last one carrying the aggregate — and the generated CreateTextToSpeechAsync tries to deserialize that SSE body as a single TextToSpeechResponse. There is no CreateTextToSpeechAsStreamAsync, so streaming synthesis is unreachable through the package and a consumer cannot get first-audio latency from MiniMax without a raw HttpClient.

Fix

Add the text/event-stream response (a TextToSpeechStreamEvent/TextToSpeechResponse per event) to the operation so AutoSDK emits the IAsyncEnumerable variant — the same shape Mistral.ChatClient.ChatCompletionAsStreamAsync already has in a sibling SDK.

Found while auditing which tryAGI TTS SDKs can feed a PCM streaming sink (HavenDV/Advantage, 2026-09-15).

Start in src/libs/MiniMax/openapi.yaml at the createTextToSpeech operation and review TtsStreamOptions plus the sibling Mistral.ChatClient.ChatCompletionAsStreamAsync shape. Model the text/event-stream response so AutoSDK can emit the streaming variant, then verify that CreateTextToSpeechAsStreamAsync exposes each audio event and the final aggregate.