docs: trim the model card

feat: v1.1 base model tuned on production calls; v1 files kept with -v1 suffix

docs: production-call results and timings

docs: trained from Whisper weights with the Smart Turn recipe, not from Smart Turn v3.2

updates README

docs: correct the training-data claim, tighten the intro, add single and batch ONNX examples

docs: explain the accuracy-versus-AUC inversion in plain terms

docs: explain threshold-driven inversions; AUC on every chart row

docs: rewrite the introduction

docs: SVG charts, timings, public release

docs: restructure model card, pair groups with same-precision Smart Turn v3.2 files, add interactive charts

feat: Indic Smart Turn base/tiny in fp32 and int8 with model card, charts and evaluation reports

initial commit