CoEdIT Base LiteRT INT8 conversion

This distribution is a converted and quantized form of jbochi/coedit-base:
https://huggingface.co/jbochi/coedit-base

The source model is a fine-tuned version of google/flan-t5-base on the CoEdIT
dataset. The conversion uses separate LiteRT encoder and decoder graphs with
INT8 dynamic-range quantization. It is not an official release from the
upstream model author, Google, or Grammarly.