fix(chat): use native async LLM call to prevent session reset on message submit

The previous implementation used asyncio.to_thread(llm.complete) to avoid
blocking the event loop, but Chainlit's contextvars context is not reliably
propagated across thread boundaries, causing the session to reset and clear
the message history on each user submission.

Changes:
- Add llm.acomplete() using litellm.acompletion() (native coroutine)
- Replace asyncio.to_thread() in on_message with await llm.acomplete()
- Store the welcome message in state.messages so it is replayed on reconnect
- Persist state to cl.user_session immediately after appending the user's
  message (before the LLM call) so reconnect detection has the latest history
- Add pytest-asyncio dev dependency and asyncio_mode = "auto" config
- Add 6 async tests for acomplete() in tests/test_llm.py

https://claude.ai/code/session_01SUWzMzFvSfWiHXA2p6rPg9

tests/test_llm.py

@@ -6,6 +6,17 @@ from src import llm
 from src.models.conversation import ChatMessage
 
 
+def _make_async_mock(mocker, content: str):
+    """Return an awaitable mock that resolves to a response with the given content."""
+    mock_response = mocker.MagicMock()
+    mock_response.choices[0].message.content = content
+
+    async def _coro(*args, **kwargs):
+        return mock_response
+
+    return mocker.patch("litellm.acompletion", side_effect=_coro), mock_response
+
+
 class TestLlmComplete:
     def test_returns_model_reply(self, mocker):
         mock_response = mocker.MagicMock()
@@ -122,3 +133,63 @@ class TestLlmStreamComplete:
 
         messages = mock_completion.call_args.kwargs["messages"]
         assert messages[1] == {"role": "user", "content": "Hallo"}
+
+
+class TestLlmAComplete:
+    @pytest.mark.asyncio
+    async def test_returns_model_reply(self, mocker):
+        mock_acompletion, _ = _make_async_mock(mocker, "Hallo! Wie heisst dein Kind?")
+
+        result = await llm.acomplete("anthropic/claude-opus-4-6", "system prompt", [])
+
+        assert result == "Hallo! Wie heisst dein Kind?"
+
+    @pytest.mark.asyncio
+    async def test_passes_model_to_litellm(self, mocker):
+        mock_acompletion, _ = _make_async_mock(mocker, "ok")
+
+        await llm.acomplete("openai/gpt-4o", "system", [])
+
+        call_kwargs = mock_acompletion.call_args.kwargs
+        assert call_kwargs["model"] == "openai/gpt-4o"
+
+    @pytest.mark.asyncio
+    async def test_system_prompt_prepended_as_system_message(self, mocker):
+        mock_acompletion, _ = _make_async_mock(mocker, "ok")
+
+        await llm.acomplete("anthropic/claude-opus-4-6", "You are helpful.", [])
+
+        messages = mock_acompletion.call_args.kwargs["messages"]
+        assert messages[0] == {"role": "system", "content": "You are helpful."}
+
+    @pytest.mark.asyncio
+    async def test_chat_messages_appended_after_system(self, mocker):
+        mock_acompletion, _ = _make_async_mock(mocker, "ok")
+
+        chat = [
+            ChatMessage(role="user", content="Hallo"),
+            ChatMessage(role="assistant", content="Guten Tag"),
+        ]
+        await llm.acomplete("anthropic/claude-opus-4-6", "system", chat)
+
+        messages = mock_acompletion.call_args.kwargs["messages"]
+        assert messages[1] == {"role": "user", "content": "Hallo"}
+        assert messages[2] == {"role": "assistant", "content": "Guten Tag"}
+
+    @pytest.mark.asyncio
+    async def test_max_tokens_passed(self, mocker):
+        mock_acompletion, _ = _make_async_mock(mocker, "ok")
+
+        await llm.acomplete("anthropic/claude-opus-4-6", "system", [])
+
+        assert mock_acompletion.call_args.kwargs["max_tokens"] == 2048
+
+    @pytest.mark.asyncio
+    async def test_exception_propagates(self, mocker):
+        async def _raise(*args, **kwargs):
+            raise RuntimeError("API error")
+
+        mocker.patch("litellm.acompletion", side_effect=_raise)
+
+        with pytest.raises(RuntimeError, match="API error"):
+            await llm.acomplete("anthropic/claude-opus-4-6", "system", [])