docs: refresh BU Bench best-models plot and recommendations

[Alezander9]

The README best-models radial plot was stale: static/browsercode_best_models_*.png were last updated 2026-07-01 (5-model layout), while benchmark-x-laminar rebuilt it on 2026-08-04 (64ee505, 2-column layout with Luna as the hero card).

• Copied the current light/dark PNGs from benchmark-x-laminar/plotting/bu_bench/official_plots/ (hashes now match).
• Rewrote the recommended-model list, which still named the old plot: claude-opus-4-7 / gpt-5.5 / glm-5.2 / gemini-3.1-pro-preview / minimax-m3. The current plot has three tiers only: Opus 4.8 (best performance, 89.5%), Kimi K3 (best open-weight, 86.0%), Luna xhigh (best value, 82.0%).

Note: the plot labels the top card "Opus 4.8" but is rendered from claude-opus-4-7 data (deliberate relabel, benchmark-x-laminar b7868ca). The README follows the plot label.

---

Summary by cubic

Refreshed the BU Bench best-models plot and aligned README recommendations with the latest evals. Updated light/dark images and now show three picks: claude-opus-4-8 (best performance), kimi-k3 (best open-weight), gpt-5.6-luna at xhigh (best value).

• Notes
  • The plot labels claude-opus-4-8 but uses claude-opus-4-7 data; the README mirrors the label.

^{Written for commit b94c6ac. Summary will update on new commits.}