Add XNNPACK int8 (8da8w): WER within -1.0/+0.3 of fp32 on en/de/es/pl/fr/zh

Add Vulkan variants (fp16, int8)

Add int8 MLX variant alongside bf16

Add MLX model

Remove old fp32

Add CoreML model

Restructure to MODEL_SPEC.md convention

Improve exported model & bundle methods

Upload 2 files

initial commit