Test module fixes

Author: cryptopoly

backend_service/inference.py

@@ -1852,8 +1852,13 @@ def load_model(
         # particular model architecture.
         attempts: list[tuple[str, bool, bool]] = [(cache_strategy, fit_model_in_memory, False)]
         if cache_strategy not in ("native", "chaosengine"):
-            chaosengine_strat = _strategy_registry.get("chaosengine")
-            if chaosengine_strat and chaosengine_strat.is_available():
+            # Always include ChaosEngine as an intermediate fallback.  Its
+            # llama.cpp path only emits standard cache-type flags (q4_0 etc.)
+            # and runs on the standard binary — it does NOT require the
+            # chaos_engine Python package to be installed.  Gating on
+            # is_available() would skip this fallback on CI / dev machines
+            # that don't have the package, breaking the 3-level chain.
+            if _strategy_registry.get("chaosengine") is not None:
                 attempts.append(("chaosengine", False, True))
         if cache_strategy != "native":
             attempts.append(("native", False, True))

dflash/__init__.py

@@ -169,7 +169,10 @@ def is_ddtree_available() -> bool:
     DDTree requires the same dflash_mlx runtime as linear DFlash, plus
     access to ``dflash_mlx.runtime`` primitives for tree verification.
     """
-    runtime_spec = importlib.util.find_spec("dflash_mlx.runtime")
+    try:
+        runtime_spec = importlib.util.find_spec("dflash_mlx.runtime")
+    except ModuleNotFoundError:
+        return False
     if runtime_spec is None:
         return False
     runtime_path = getattr(runtime_spec, "origin", None)