Streams the LLM response as an async generator of chunks.
Streaming does not retry. If the connection fails mid-stream, partial tokens
have already been yielded to the caller, so retrying would produce duplicate or
incoherent output. Errors surface as typed AgentCraftErrors.
For retry-on-failure before the first token, use AgentPool with a fallback Agent.
Streams the LLM response as an async generator of chunks.
Streaming does not retry. If the connection fails mid-stream, partial tokens have already been yielded to the caller, so retrying would produce duplicate or incoherent output. Errors surface as typed AgentCraftErrors.
For retry-on-failure before the first token, use AgentPool with a fallback Agent.