fix(evaluator): deduplicate GT queries and tolerate minor extraneous entries

Fix evaluator warnings and album breakdown display

v2.2: opt-in leaderboard, admin panel, column selector, full-submission requirement

v2.1: remove Chinese text, Apple-style UI, centered submit form, auto-seed leaderboard data

v2: per-album eval + avg, email+rate limit, top-30, 30d cleanup, luxury UI, import paper data

Rewrite PhotoBench-Full leaderboard: drag-drop submission, on-the-fly evaluation, top-20 display with method filter