Switch to 35B i1-Q4_K_M (same compression, faster load); size GPU reservations from measurements

UI: word-hugging token shading, cold-start note, unlock controls after errors

Refine UI: typography, in-panel progress, inline errors, token tooltips

Replace mode tabs with centered swap control and preserve drafts

Use arithmetic-coding description and symmetric Compress/Restore layouts

Simplify UI, show token coding costs, and skip automatic decode verification