feat: handle response from VertexAI instead of OpenAI
Description
Since CLI receives the raw response from the model, the logic that handles the response must be changed.
The feature is in alpha stage, so it's fine to avoid backward compatibility. The backend still responds with OpenAI by default, so the older versions of CLI will still work while the newer ones will send VertexAI model as a param and receive and handle VertexAI response instead.