Update README.md

Model card: fix redundant 'vocabulary lexicon' wording

Model card: drop pipeline_tag (no HF hosted inference for .pte)

Model card: clarify I/O table labels + downsample note

Add model card: overview, models, ExecuTorch quickstart