fix: allow interrupting on each token stream

ref: N25B-452

src/control_backend/agents/llm/llm_agent.py

@@ -179,6 +179,9 @@ class LLMAgent(BaseAgent):
             full_message = ""
             current_chunk = ""
             async for token in self._stream_query_llm(messages):
+                if self._interrupted:
+                    return
+
                 full_message += token
                 current_chunk += token