Pin Core ML fp16 to CPU_AND_NE (6.0x on s, 2.8x on x)

Core ML: drop fp32, fp16 is equivalent in use

Remove MLX artifacts: slower than Core ML on device

Rebuild mlx export with fork toolchain (ExecuTorch 1.4.1)

Add backend config (ExecuTorch 1.4.1)

Add export (ExecuTorch 1.4.1)

Update models

fix(v0.9.0): switch to size-first layout (s/{xnnpack,coreml}/ + x/{xnnpack,coreml}/)