FluidInference/paraformer-large-zh-coreml

docs: fp16/int8 full benchmark + RTFx

Add ParaformerDecoder_int8 (int8, ~half size, accuracy-neutral)

Add ParaformerEncoder_int8 (int8, ~half size, accuracy-neutral)

docs: full AISHELL CER 2.12%

Add CIF alphas model (host does only integrate-and-fire)

Add Paraformer-large zh CoreML (preprocessor + encoder fp16/ANE + decoder fp16/ANE + vocab + card)

initial commit