feat(coreml): publish fp16 only, drop the fp32 build

feat(coreml): add the fp16 build, pinned to CPU_AND_GPU

feat(coreml): add the fp16 build, pinned to CPU_AND_GPU

Revert the fp16 Core ML build: it returns no detections

Add the fp16 Core ML build

Add coreml config (fp32, 576x576, 100 detections)

Restore Core ML keypoint export: verified working on iOS device

Remove Core ML keypoint export: returns no keypoints on iOS device (host numerics are exact; failure is device-side)

Restore Core ML keypoint export: prior removal was based on a faulty test (non-contiguous input tensor)

Remove Core ML keypoint export: produces no detections (max score 0.005 vs 0.89 reference)

Add per-backend config

Add per-backend config

Remove mlx export uploaded at incorrect top-level path (belongs under preview/)

Rebuild mlx export with fork toolchain (ExecuTorch 1.4.1)

Add mlx export (ExecuTorch 1.4.1)

Use iOS16 CoreML model for keypoint preview

docs: add README

chore: remove redundant file

Add RF-DETR keypoint preview models and config

initial commit