docs: the serving addon is released; point at the newer bundled model

Refresh README; document sft-c16 (latent-only reasoning, c=16, learned stop)

Add sft-c16: phase6 Fix-B closed-loop SFT head (c=16, format_version 3)

Update README.md

Update README.md

Add CoLaR reasoning head (SFT + GRPO), model card, and technical report

initial commit