fix(agent): deliver LLM errors to streaming channels and avoid polluting session context

When the LLM returns an error (e.g. 429 quota exceeded, stream timeout),
streaming channels silently drop the error message because `_streamed=True`
is set in metadata even though no content was actually streamed.

This change:
- Skips setting `_streamed` when stop_reason is "error", so error messages
  go through the normal channel.send() path and reach the user
- Stops appending error content to session history, preventing error
  messages from polluting subsequent conversation context
- Exposes stop_reason from _run_agent_loop to enable the above check

tests/agent/test_runner.py

@@ -798,7 +798,7 @@ async def test_loop_max_iterations_message_stays_stable(tmp_path):
     loop.tools.execute = AsyncMock(return_value="ok")
     loop.max_iterations = 2
 
-    final_content, _, _ = await loop._run_agent_loop([])
+    final_content, _, _, _ = await loop._run_agent_loop([])
 
     assert final_content == (
         "I reached the maximum number of tool call iterations (2) "
@@ -825,7 +825,7 @@ async def test_loop_stream_filter_handles_think_only_prefix_without_crashing(tmp
     async def on_stream_end(*, resuming: bool = False) -> None:
         endings.append(resuming)
 
-    final_content, _, _ = await loop._run_agent_loop(
+    final_content, _, _, _ = await loop._run_agent_loop(
         [],
         on_stream=on_stream,
         on_stream_end=on_stream_end,
@@ -849,7 +849,7 @@ async def test_loop_retries_think_only_final_response(tmp_path):
 
     loop.provider.chat_with_retry = chat_with_retry
 
-    final_content, _, _ = await loop._run_agent_loop([])
+    final_content, _, _, _ = await loop._run_agent_loop([])
 
     assert final_content == "Recovered answer"
     assert call_count["n"] == 2