Perf: max_new 120->80 for free CPU tier

Perf: cut max_new_tokens 200->120; honest spinner + README about CPU latency

Fix prompt-text-area sync bug; expand citations (CAA + Persona Vectors + Arditi)

Fix: empty-state message no longer says at-least-one knob

Validate demo prompts on 1.5B; replace 2 weak flips with stronger ones

UX: prompt buttons show topic emoji + preview

UX: prompt always on top, only matching knob clickable

Initial deploy: Hidden Directions demo (Qwen 2.5-1.5B-Instruct)

initial commit