Stop restating quantized, and drop the unused default

perf(coreml): run decode on the CPU, not the Neural Engine

perf(coreml): run decode on the CPU, not the Neural Engine

Update .pte files for v0.10.0

Replace .pte files with reexported versions including dynamic dims companion methods

Fix wrong metas in coreml/config.json

Remove old fp32 entry

Check the fp16 model

fix(v0.9.0): add missing coreml/config.json

Add CoreML model