feat: capture streaming token usage in QwenClient and QwenVLClient

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

backend/tests/observability/test_stream_chat_usage_capture.py

@@ -81,3 +81,34 @@ def test_deepseek_stream_chat_without_usage_chunk_returns_none():
 
     assert chunks == ["Hi"]
     assert returned_usage is None
+
+
+from app.services.llm.qwen_client import QwenClient, QwenVLClient
+
+
+def test_qwen_stream_chat_returns_usage_from_trailing_chunk():
+    """QwenClient.stream_chat() must return the trailing usage dict."""
+    config = LLMConfig(provider=LLMProvider.QWEN, model="qwen3.5-flash", api_key="k", base_url="http://x/v1")
+    client = QwenClient(config)
+    usage = {"prompt_tokens": 10, "completion_tokens": 4, "total_tokens": 14}
+    client._client = _mock_streaming_client(_sse_lines("Bonjour", usage=usage))
+
+    chunks, returned_usage = _drain(client.stream_chat([{"role": "user", "content": "hi"}]))
+
+    assert chunks == ["Bonjour"]
+    assert returned_usage == usage
+    sent_payload = client._client.stream.call_args.kwargs["json"]
+    assert sent_payload["stream_options"] == {"include_usage": True}
+
+
+def test_qwen_vl_stream_chat_returns_usage_from_trailing_chunk():
+    """QwenVLClient.stream_chat() must return the trailing usage dict."""
+    config = LLMConfig(provider=LLMProvider.QWEN_VL, model="qwen3-vl-plus", api_key="k", base_url="http://x/v1")
+    client = QwenVLClient(config)
+    usage = {"prompt_tokens": 20, "completion_tokens": 6, "total_tokens": 26}
+    client._client = _mock_streaming_client(_sse_lines("Describing image", usage=usage))
+
+    chunks, returned_usage = _drain(client.stream_chat([{"role": "user", "content": "describe"}]))
+
+    assert chunks == ["Describing image"]
+    assert returned_usage == usage