feat: ONNX Runtime + CoreML EP for YOLO detection on Apple Silicon

- Ship yolo26n.onnx (9.4MB) from onnx-community for zero-config default
- On-demand download for s/m/l from onnx-community HuggingFace repos
- _OnnxCoreMLModel wrapper: logits+pred_boxes → sigmoid+cxcywh→xyxy
- Bypasses macOS 26.x MPSGraph MLIR crash (SIGABRT)
- 15-25ms/frame (~67 FPS) on Apple M5 Pro with CoreML EP
- Strip requirements_mps.txt: no torch/ultralytics (~120MB → ~17MB)
- Class names from yolo26n_names.json (no .pt dependency)

Author: solderzzc

skills/detection/yolo-detection-2026/requirements_mps.txt

@@ -1,10 +1,8 @@
 # YOLO 2026 — MPS (Apple Silicon) requirements
-# Standard PyTorch — MPS backend is included by default on macOS
-torch>=2.4.0
-torchvision>=0.19.0
-ultralytics>=8.3.0
-coremltools>=8.0
+# Uses ONNX Runtime + CoreML EP for GPU/ANE acceleration.
+# Pre-built yolo26n.onnx is shipped in the repo, so torch/ultralytics
+# are NOT needed at runtime — only onnxruntime for inference.
+onnxruntime>=1.19.0
 numpy>=1.24.0,<2.0.0
 opencv-python-headless>=4.8.0
 Pillow>=10.0.0
-