Add per-metric good/bad video examples

- New Qualitative Examples section, placed first, grouped into Action
  Dynamics / World Memory / Visual Quality with per-group tabs
- 18 clips (9 metrics x high/low), same scene and command within a pair,
  re-encoded to 576w CRF28 (~38 MB total incl. posters)
- Only the visible pair loads and plays; others stay paused via
  IntersectionObserver, so idle groups cost no bandwidth
- Remove the benchmark comparison table and the hero stat cards
- Tighten Why WorldMark bullets
- Anchor eval-output ignores to repo root so docs/static/videos is tracked

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: xjxu21

.gitignore

@@ -17,10 +17,10 @@ worldmark_alaya/
 worldmark_alaya.zip
 
 # Evaluation outputs
-results/
-results_*/
-videos/
-logs/
+/results/
+/results_*/
+/videos/
+/logs/
 *.npz
 worldmark_code/
 worldmark_code.zip