Fix issue #5349: Sub-agent with sequential LRO tools fails to resume.

Modified `_run_async_impl` in `llm_agent.py` to prevent premature `end_of_agent` marking when a sub-agent pauses for an LRO. Added `sub_agent_paused` tracking to ensure `end_of_agent=True` is only set if the sub-agent actually finished.

Author: 9chait9

src/google/adk/agents/llm_agent.py

@@ -474,13 +474,15 @@ async def _run_async_impl(
     if agent_state is not None and (
         agent_to_transfer := self._get_subagent_to_resume(ctx)
     ):
-      sub_agent_paused = False
+      sub_agent_paused = False  # Track pause state
       async with Aclosing(agent_to_transfer.run_async(ctx)) as agen:
         async for event in agen:
+          # Check if this event signals a pause (requires Bug 1 fix context)
           if ctx.should_pause_invocation(event):
             sub_agent_paused = True
           yield event
 
+      # Only mark as ended if it didn't pause
       if not sub_agent_paused:
         ctx.set_agent_state(self.name, end_of_agent=True)
         yield self._create_agent_state_event(ctx)