Apple Silicon + Windows CUDA perf: 60 FPS pipeline, cross-platform routing

Bundles CoreML graph rewrites, GPU-accelerated pipeline work, Windows CUDA
fixes, and Mac/Windows runtime routing into a single drop.

CoreML (Apple Silicon):
- Decompose Pad(reflect) → Slice+Concat in inswapper_128 so the model
  runs in one CoreML partition instead of 14 (TEMPORARY: fixed upstream
  in microsoft/onnxruntime#28073, drop when ORT >= 1.26.0).
- Fold Shape/Gather chains to constants in det_10g (21ms → 4ms).
- Decompose Split(axis=1) → Slice pairs in GFPGAN (155ms → 89ms).
- Route detection model to GPU so the ANE is free for the swap model.
- Centralize provider/config selection in create_onnx_session.

Pipeline (all platforms):
- Parallelize face landmark + recognition post-detection; skip landmark_2d_106
  when only face_swapper is active.
- Pipeline face detection with swap for ANE overlap.
- GPU-accelerated paste_back, MJPEG capture, zero-copy display path.
- Standalone pipeline benchmark script.

Windows / CUDA:
- CUDA graphs + FP16 model + all-GPU pipeline for 1080p 60 FPS.
- Auto-detect GPU provider and fix DLL discovery for Windows CUDA execution.

Cross-platform:
- platform_info helper for Mac/Windows runtime routing.
- GFPGAN 30 fps + MSMF camera 60 fps with adaptive pipeline tuning.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

modules/gpu_processing.py

@@ -18,6 +18,7 @@ Usage
 
 from __future__ import annotations
 
+import os
 import cv2
 import numpy as np
 from typing import Tuple, Optional
@@ -27,20 +28,25 @@ from typing import Tuple, Optional
 # ---------------------------------------------------------------------------
 CUDA_AVAILABLE: bool = False
 
-try:
-    # cv2.cuda.GpuMat is only present when OpenCV is compiled with CUDA
-    _test_mat = cv2.cuda.GpuMat()
-    # Verify we have the required filter / image-processing functions
-    _has_gauss = hasattr(cv2.cuda, "createGaussianFilter")
-    _has_resize = hasattr(cv2.cuda, "resize")
-    _has_cvt = hasattr(cv2.cuda, "cvtColor")
-    if _has_gauss and _has_resize and _has_cvt:
-        CUDA_AVAILABLE = True
-        print("[gpu_processing] OpenCV CUDA support detected – GPU-accelerated processing enabled.")
-    else:
-        pass  # silently fall back to CPU
-except Exception:
-    pass  # silently fall back to CPU
+# OpenCV CUDA per-operation acceleration is DISABLED by default.
+# Each gpu_* call uploads to GPU, processes, then downloads back to CPU.
+# At webcam resolution (~960x540) this upload/download overhead far exceeds
+# the time saved on the actual operation, making it slower than pure CPU.
+# The heavy lifting (face detection, swap, enhancement) runs on GPU via
+# ONNX Runtime's CUDAExecutionProvider, which is where GPU matters.
+#
+# To force-enable, set OPENCV_CUDA_PROCESSING=1 in your environment.
+if os.environ.get("OPENCV_CUDA_PROCESSING") == "1":
+    try:
+        _test_mat = cv2.cuda.GpuMat()
+        _has_gauss = hasattr(cv2.cuda, "createGaussianFilter")
+        _has_resize = hasattr(cv2.cuda, "resize")
+        _has_cvt = hasattr(cv2.cuda, "cvtColor")
+        if _has_gauss and _has_resize and _has_cvt:
+            CUDA_AVAILABLE = True
+            print("[gpu_processing] OpenCV CUDA processing enabled via OPENCV_CUDA_PROCESSING=1.")
+    except Exception:
+        pass
 
 
 # ---------------------------------------------------------------------------