[feat] Output streaming

[bboynton97]

This is a pretty big feature idea with large implications

Using an API based agent can be frustrating because of how much time it takes between sending the initial input and receiving the output.

Especially with deployed agents, it would be a good idea to be able to send a webhook that can receive streamed updates. It should also send event types (thoughts, tool use, final result) for at least each llm call. Or even better, stream tokens as generated.

This is a long-term roadmap vision issue