test: pin nemotron fallback to nim path

tests/test_nemotron_fallback.py

@@ -1,5 +1,18 @@
 from dataclasses import dataclass
 
+import pytest
+
+
+@pytest.fixture(autouse=True)
+def _force_legacy_nim_first(monkeypatch):
+    """Fallback tests cover the legacy NIM path, not live qwen3/Ollama routing."""
+    import services.nemoton_dispatcher_service as module
+
+    module._ALERT_CACHE.clear()
+    monkeypatch.setattr(module, "NEMOTRON_OLLAMA_FIRST", False)
+    yield
+    module._ALERT_CACHE.clear()
+
 
 @dataclass
 class FakeThreat: