Support thinking blocks for agents during orchestration steps #642
Description
Today, we only support send back content_type: text back to the client when in the process of multi-agent orchestration. This is limiting because the user doesn't see forward progress. We should add support for "thinking" blocks that are sent from an agent upstream to the client and let the client render those appropriately to the user.
Btw, v1/responses and v1/message support thinking blocks (v1/chat/completions does not)