Fix README: correct quantization description, remove mlx-audio version reference

Fix: save weights without model. prefix so apply_quantization predicate matches correctly

Add 8-bit quantized fish-audio-s2-pro (group_size=64, bits=8)

initial commit