Playgrounds: launch eval and quick figures in Personal threads #511

Tracks R1: quick-figure mode and the 20-prompt launch eval.

Done when: on the 20-prompt launch eval, run in fresh Personal threads on the default Claude Code and Codex models:

• ≥18/20 cards have every control live, checked by scripts/drive.mjs
• 0 clarifying questions
• median ≤3 min and p90 ≤6 min from send to a visible card
• progress shows while the figure builds
• it works in a Personal thread with no Git checkout

Also covers the CDN "explain to my team" stall from #460.

PR: #508

BB-Thread-ID: thr_vv8zgxmczt