[Feature] Streaming output for agent execution

## Summary Support real-time streaming output during agent execution, so users can see LLM responses as they are generated. ## Motivation Currently `AgentRunner` uses `adapter.chat()` which waits for the full response. For long-running tasks, users have no visibility into what the agent is doing until it finishes. Streaming would enable: - Real-time progress feedback in CLI or web UI - Lower perceived latency - Early termination if the agent goes off track ## Proposed Approach - Add `stream` mode option to `AgentRunner.run()` - Use `adapter.stream()` instead of `adapter.chat()` when enabled - Emit events via callback or AsyncIterable for consumer integration - Handle tool calls within streaming context ## Open Questions - Should streaming be opt-in per agent or per team? - How to handle tool execution interleaved with streaming output?