The streamed chunks are correct: <img width="1405" height="993" alt="Image" src="https://github.com/user-attachments/assets/4c0bd741-8cfe-443d-a7df-c5d19c7544aa" /> But the final result (with aggregated parts) does not correctly convert thinking parts to Genkit reasoning parts. <img width="1400" height="978" alt="Image" src="https://github.com/user-attachments/assets/07074a66-5420-43cc-9b7e-52e8b6dd5883" /> This is reproducible with https://github.com/firebase/genkit/pull/3179