test(ollama): align inference connect errors with degraded health

apps/api/tests/test_ollama_health_monitor.py

@@ -243,8 +243,8 @@ class TestInference:
         assert report.status == HealthStatus.DEGRADED
 
     @pytest.mark.asyncio
-    async def test_inference_connect_error_offline(self, monitor):
-        """推理 ConnectError → OFFLINE（連通性層已放行，但推理層掉線）"""
+    async def test_inference_connect_error_degraded(self, monitor):
+        """推理 ConnectError → DEGRADED（連通性已通過，視為 socket 瞬斷）"""
         mock_client = AsyncMock()
         mock_client.post = AsyncMock(side_effect=httpx.ConnectError("refused"))
         mock_client.__aenter__ = AsyncMock(return_value=mock_client)
@@ -253,7 +253,7 @@ class TestInference:
         with patch("httpx.AsyncClient", return_value=mock_client):
             report = await monitor._check_inference(HOST)
 
-        assert report.status == HealthStatus.OFFLINE
+        assert report.status == HealthStatus.DEGRADED
 
     @pytest.mark.asyncio
     async def test_inference_non_200_degraded(self, monitor):