Embed evaluation card in Summary reader mode by default

Tell agent recency is prioritized; expose newest_model_date

Rank search by benchmark recency (newest model on leaderboard)

Redesign UI to match EvalEval/evalcards design system (paper-aligned tokens, Inter/IBM Plex Mono, brand blue)

Fix: handle None constraints/max_params from Gradio inputs

Best Model Finder: agentic search over the EEE datastore

initial commit