When streaming text through the ai sdk using openai's o3-mini, there are no chunks for reasoning, despite there being reasoning tokens used at the end as evident by event.experimental_providerMetadata.

const result = streamText({
  model: registry.languageModel(`openai:${o3m}`),
  prompt: message.payload.message,
  maxSteps: 10,
  onChunk(event) {
    console.log("onChunk event.type", event.chunk.type);
  },
  onFinish: (event) => {
    console.log("onFinish event.reasoning", event.reasoning);
    console.log("onFinish providerMetadata", event.experimental_providerMetadata);
  },
});

const dataStreamResponse = result.toDataStreamResponse({
  sendReasoning: true,
});