AD-Styles/mini-llava-v3

docs: drop "(정직한 명시)" subtitle from §변하지 않은 것

docs: 3-way intro (한국어 재학습 / 환각 wrapper+OOD / 배포 Slim adapter)

docs: portfolio-tone rebalance (drop self-deprecating PEFT callout, neutral Slim heading)

docs: soften OOD claim, name slim as PEFT workaround, clarify VQA scope

docs: sync model card with corrected VQAv2 measurement (+2pp v3>v2)

Honest messaging: separate capability gains (Korean, OOD) from deployment optimization (Slim)

Add v3 slim adapter (1045MB → 8.28MB, no quality loss)

Add v3 projector (Korean training, 175 min)

Add v3 model card (Korean + Slim + OOD)

initial commit