Add reviewed agent decision demos and LLM-authored WebShop menus

[weitianxin]

Summary

• make WebShop candidate groups LLM-authored and keep completion with the frontier LLM
• add paired WebShop supplement plus reviewed 1080p WebShop, FrozenLake, and Terminal-Bench decision animations
• integrate task/delegation principles, results, and demos into both READMEs and the technical report

Validation

• independent decision-demo reviewer: PASS on all seven criteria
• 52 focused harness tests pass
• WebShop paired supplement: 3 seeds; success 67% to 100%, calls -21.4%, tokens -14.3%, time -15.0%