fix: resolve ONNX CUDA fallback, Docker e2e infrastructure, and all test failures

- Add safe_onnx_session() to gpu.py with graceful CUDA EP → CPU fallback
- Replace bare ort.InferenceSession() calls across colorize, restore, inpaint, remove_bg
- Add libcublas-12-6 to production Dockerfile for ONNX Runtime CUDA EP
- Add skipIfFeatureNotInstalled guards to remove-bg, blur-faces, smart-crop, ocr, noise-removal e2e specs
- Add AI tool install prompt detection in tools-all.spec.ts
- Add smart-crop to PYTHON_SIDECAR_TOOLS so frontend shows install prompt correctly
- Create Dockerfile.test.dockerignore to include tests/ in test image builds
- Add libheif-examples and exiftool to Dockerfile.test for HEIC and metadata tests
- Regenerate visual regression baselines for Docker/Linux and skip on non-Docker platforms

packages/ai/python/colorize.py

@@ -46,19 +46,11 @@ OPENCV_POINTS_PATH = os.environ.get(
 
 def colorize_ddcolor(img_bgr, intensity):
     """Colorize using DDColor ONNX model."""
-    import onnxruntime as ort
+    from gpu import safe_onnx_session
 
     emit_progress(15, "Loading DDColor model")
 
-    providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
-    try:
-        from gpu import gpu_available
-        if not gpu_available():
-            providers = ["CPUExecutionProvider"]
-    except ImportError:
-        providers = ["CPUExecutionProvider"]
-
-    session = ort.InferenceSession(DDCOLOR_MODEL_PATH, providers=providers)
+    session = safe_onnx_session(DDCOLOR_MODEL_PATH)
     input_name = session.get_inputs()[0].name
     input_shape = session.get_inputs()[0].shape
     # Dynamic dims are strings ('w', 'h'), so default to 512 if not int

packages/ai/python/gpu.py

@@ -55,3 +55,20 @@ def onnx_providers():
     if gpu_available():
         return ["CUDAExecutionProvider", "CPUExecutionProvider"]
     return ["CPUExecutionProvider"]
+
+
+def safe_onnx_session(model_path, providers=None):
+    """Create an ONNX Runtime InferenceSession with graceful CUDA EP fallback."""
+    import onnxruntime as ort
+
+    if providers is None:
+        providers = onnx_providers()
+
+    try:
+        return ort.InferenceSession(model_path, providers=providers)
+    except Exception as e:
+        if "CUDAExecutionProvider" in providers:
+            print(f"[gpu] CUDA EP init failed ({e}), falling back to CPU",
+                  file=sys.stderr, flush=True)
+            return ort.InferenceSession(model_path, providers=["CPUExecutionProvider"])
+        raise

packages/ai/python/inpaint.py

@@ -99,7 +99,7 @@ def main():
 
         try:
             import cv2
-            import onnxruntime as ort
+            import onnxruntime
         except ImportError as e:
             print(json.dumps({
                 "success": False,
@@ -110,11 +110,8 @@ def main():
         emit_progress(10, "Loading model")
         model_path = _get_model_path()
 
-        # Configure ONNX Runtime session
-        from gpu import onnx_providers
-        providers = onnx_providers()
-
-        session = ort.InferenceSession(model_path, providers=providers)
+        from gpu import safe_onnx_session
+        session = safe_onnx_session(model_path)
 
         emit_progress(20, "Loading images")
         img = Image.open(input_path).convert("RGB")

packages/ai/python/remove_bg.py

@@ -76,7 +76,16 @@ def main():
 
         emit_progress(10, "Loading model")
 
-        session = new_session(model, providers=onnx_providers())
+        providers = onnx_providers()
+        try:
+            session = new_session(model, providers=providers)
+        except Exception as e:
+            if "CUDAExecutionProvider" in providers:
+                print(f"[remove-bg] GPU session failed ({e}), falling back to CPU",
+                      file=sys.stderr, flush=True)
+                session = new_session(model, providers=["CPUExecutionProvider"])
+            else:
+                raise
 
         emit_progress(25, "Model loaded")
 

packages/ai/python/restore.py

@@ -152,14 +152,10 @@ def inpaint_damage(img_bgr, mask):
     Returns:
         Restored BGR image with damage inpainted.
     """
-    import onnxruntime as ort
+    from gpu import safe_onnx_session
 
     model_path = _get_lama_path()
-    providers = ["CPUExecutionProvider"]
-    if "CUDAExecutionProvider" in ort.get_available_providers():
-        providers.insert(0, "CUDAExecutionProvider")
-
-    session = ort.InferenceSession(model_path, providers=providers)
+    session = safe_onnx_session(model_path)
 
     orig_h, orig_w = img_bgr.shape[:2]
     img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
@@ -261,7 +257,7 @@ def enhance_faces(img_bgr, fidelity=0.7):
         Tuple of (enhanced BGR image, number of faces found).
     """
     import mediapipe as mp
-    import onnxruntime as ort
+    from gpu import safe_onnx_session
 
     # Detect faces
     img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
@@ -321,11 +317,7 @@ def enhance_faces(img_bgr, fidelity=0.7):
 
     # Load CodeFormer model
     model_path = _get_codeformer_path()
-    providers = ["CPUExecutionProvider"]
-    if "CUDAExecutionProvider" in ort.get_available_providers():
-        providers.insert(0, "CUDAExecutionProvider")
-
-    session = ort.InferenceSession(model_path, providers=providers)
+    session = safe_onnx_session(model_path)
     input_names = [inp.name for inp in session.get_inputs()]
 
     result = img_bgr.copy()
@@ -476,20 +468,12 @@ def colorize_bw(img_bgr, intensity=0.85):
 
     Reuses the DDColor model that the colorize tool already downloads.
     """
-    import onnxruntime as ort
+    from gpu import safe_onnx_session
 
     if not os.path.exists(DDCOLOR_MODEL_PATH):
         return img_bgr, False
 
-    providers = ["CPUExecutionProvider"]
-    try:
-        from gpu import gpu_available
-        if gpu_available():
-            providers.insert(0, "CUDAExecutionProvider")
-    except ImportError as e:
-        print(f"[restore] GPU detection unavailable: {e}", file=sys.stderr, flush=True)
-
-    session = ort.InferenceSession(DDCOLOR_MODEL_PATH, providers=providers)
+    session = safe_onnx_session(DDCOLOR_MODEL_PATH)
     input_name = session.get_inputs()[0].name
     input_shape = session.get_inputs()[0].shape
     model_size = (