Model Builder: Add Post processing script to convert fp16/32 LM_HEAD to int8 and use tied embeddings

[sushraja-msft]

With previous change

#1436

one can produce fp16/fp32 lm_head onnx models.

With this additional post processing script such models can be converted to int8 lm_head and also use tied embeddings to save disk space.

Usage:

python PostProcess_LM_HEAD_int8_quantization_and_tied_embeddings.py phi4_fp16_lm_head\model.onnx C:\AI\final_model\model.onnx --block_size 32 --accuracy_level 4 --fp16 --tie_embeddings