feat: record streaming token usage in TrackedLLMClient.stream_chat

Implement manual generator driving using next()/StopIteration to capture
the return value (trailing usage dict) from inner stream_chat() implementations,
enabling token tracking for streaming LLM calls.

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

backend/tests/observability/test_tracked_client.py

@@ -83,3 +83,23 @@ def test_stream_chat_records_call_without_token_usage():
     entry = tracker.get("deepseek", "deepseek-v4-flash")
     assert entry.call_count_ok == 1
     assert entry.total_tokens == 0
+
+
+def test_stream_chat_records_usage_from_generator_return_value():
+    """stream_chat() must forward the inner generator's returned usage dict to record()."""
+    inner = _make_inner()
+
+    def fake_stream(*args, **kwargs):
+        yield "chunk-1"
+        yield "chunk-2"
+        return {"prompt_tokens": 6, "completion_tokens": 2, "total_tokens": 8}
+
+    inner.stream_chat.side_effect = fake_stream
+    tracker = ModelUsageTracker()
+
+    chunks = list(TrackedLLMClient(inner, tracker).stream_chat([{"role": "user", "content": "hi"}]))
+
+    assert chunks == ["chunk-1", "chunk-2"]
+    entry = tracker.get("deepseek", "deepseek-v4-flash")
+    assert entry.total_tokens == 8
+    assert entry.call_count_ok == 1