docs: add Usage section (per-task prompts + inference flags)

model card: revert description expansion

model card: capabilities, usage, limitations

model card: badge label spacing

model card: banner image + badges

model card: slim company links

model card: KoreaDeep intro + DEEP Agent banner

model card: condense serving and methodology sections

model card: neutral serving alias

model card: trim lineage details

KDL-Frontier-Parser-nano: weights + model card + ParseBench eval results (76.48 full-set single-pass)