Make the first EvalArc review clear and reproducible

[noteflowai]

EvalArc introduced several workflows before explaining how a new user could review one real failure. Lead the English/Chinese README, website and HF card with the recorded 90% → 93.75% regression, then connect that example to a verified local review using the published wheel and original evidence.

• Add a 30-second walkthrough made from asserted UI states, with captions, explicit scripted-control scope, reduced-motion fallback and optional video playback.
• Document installation outside a checkout, comparison and suite exit codes, and the bounded AgentCore export contract. Add a first-use feedback form and focused existing-channel outreach drafts.
• Remove accumulated version announcements and stale counts from current entry points. Preserve the Python version, grading semantics and released evidence.

Validation: installed the public 0.12.1 wheel in a fresh environment outside the checkout; verified eight expected command outcomes with PATH restricted to that environment. Eight site tests, lint, formatting and desktop/390px/320px browser checks passed, including decoded video playback, navigation, archive downloads and original evidence workflows. Full CI remains the merge gate.

The scored trace examples remain synthetic; this change does not establish live AgentCore integration or independent adoption.