refactor(stream): 将不完整流 token 估算逻辑收敛到 StreamContext

- 新增 has_partial_response / ensure_estimated_output_tokens / should_estimate_incomplete_tokens 方法
- CancelledError 路径在归因前即补充 output_tokens，确保日志包含估算值
- CLI Handler 和 Chat Handler 的兜底估算统一使用 should_estimate_incomplete_tokens
- 移除 cli_monitor_mixin 和 stream_telemetry 中重复的条件判断
- 新增对应单元测试

tests/api/handlers/base/test_cli_monitor_mixin.py

@@ -33,6 +33,12 @@ async def _cancel_immediately() -> AsyncGenerator[bytes, None]:
     raise asyncio.CancelledError()
 
 
+async def _yield_once_then_cancel(ctx: StreamContext) -> AsyncGenerator[bytes, None]:
+    ctx.append_text("partial output")
+    yield b"data: chunk\n\n"
+    raise asyncio.CancelledError()
+
+
 @pytest.mark.asyncio
 async def test_create_monitored_stream_marks_client_disconnected_when_confirmed() -> None:
     monitor = _DummyMonitor()
@@ -70,7 +76,9 @@ async def test_create_monitored_stream_marks_server_cancelled_when_confirmed_con
 
 
 @pytest.mark.asyncio
-async def test_create_monitored_stream_marks_cancelled_unknown_when_disconnect_check_uncertain() -> None:
+async def test_create_monitored_stream_marks_cancelled_unknown_when_disconnect_check_uncertain() -> (
+    None
+):
     monitor = _DummyMonitor()
     monitor.CANCEL_DISCONNECT_RETRY_DELAYS_SECONDS = ()
     ctx = StreamContext(model="test-model", api_format="openai:cli", request_id="req-unknown")
@@ -85,3 +93,23 @@ async def test_create_monitored_stream_marks_cancelled_unknown_when_disconnect_c
     assert ctx.status_code == 503
     assert ctx.error_message == "cancelled_unknown"
     assert "cancel_origin=cancelled_unknown" in (ctx.upstream_response or "")
+
+
+@pytest.mark.asyncio
+async def test_create_monitored_stream_estimates_output_tokens_before_unknown_cancel_log() -> None:
+    monitor = _DummyMonitor()
+    monitor.CANCEL_DISCONNECT_RETRY_DELAYS_SECONDS = ()
+    ctx = StreamContext(model="test-model", api_format="openai:cli", request_id="req-estimate")
+
+    request = _RequestStub([asyncio.TimeoutError()])
+    monitored = monitor._create_monitored_stream(ctx, _yield_once_then_cancel(ctx), request)
+
+    with pytest.raises(asyncio.CancelledError):
+        async for _ in monitored:
+            pass
+
+    expected_output_tokens = max(1, len("partial output") // 4)
+    assert ctx.status_code == 503
+    assert ctx.error_message == "cancelled_unknown"
+    assert ctx.output_tokens == expected_output_tokens
+    assert f"output_tokens={expected_output_tokens}" in (ctx.upstream_response or "")