Declare the int8 Core ML variant

Add int8 Core ML variant (embedding table quantized)

Core ML: accept a sequence length of 1

Stop restating quantized, and drop the unused default

Core ML: pin compute_unit=CPU_ONLY, drop the fp32 variant

Add per-backend config

Add coreml export (ExecuTorch 1.4.1)

Add coreml export (ExecuTorch 1.4.1)