[ML] Address Copilot review on kill switch test

- Update stale branch references to generic requirements
- Treat model generation failures as test failures, not skips —
  for security regression tests, silently skipping is unsafe

Made-with: Cursor

Author: edsavage

test/test_pytorch_inference_evil_models.py

@@ -21,8 +21,9 @@
     --binary PATH   Explicit path to the pytorch_inference binary.
                     If omitted, the script searches standard build locations.
 
-Requires: torch, a built pytorch_inference binary with graph validation
-          (feature/harden_pytorch_inference branch or later).
+Requires: torch, and a built pytorch_inference binary from this repository
+          with graph validation enabled (i.e., including the
+          CModelGraphValidator checks).
 """
 
 import argparse
@@ -215,7 +216,7 @@ def find_pytorch_inference() -> str:
 
     raise FileNotFoundError(
         "Could not find pytorch_inference binary. "
-        "Build from the feature/harden_pytorch_inference branch, or pass --binary."
+        "Build the project with graph validation enabled, or pass --binary."
     )
 
 
@@ -315,7 +316,8 @@ def run_tests(binary: str) -> bool:
                 generate_model(spec["class"], model_path)
                 print(f"  Model generated: {model_path.name} ({model_path.stat().st_size} bytes)")
             except Exception as e:
-                print(f"  SKIP: could not generate model: {e}")
+                print(f"  FAIL: could not generate model: {e}")
+                all_passed = False
                 print()
                 continue