docs: sharpen VES README positioning and onboarding

[Zhuchen00123]

What changed

• Rewrote the English and Chinese READMEs around the verifier-first story instead of leading with API inventory.
• Put the core line — “AI can propose. It cannot grade itself.” — above the fold.
• Added a concrete trust-boundary diagram, 30-second adversarial/search demo, and an eval.py comparison.
• Explained the Evidence → Judge separation with a minimal example.
• Reorganized the public API into the four frozen v0.1 layers.
• Made verify/replay, SearchEngine, security guarantees, current limitations, and v0.1 status easier to scan.
• Added a clear “Where this is going / VES Modeling” section without changing Core scope or overpromising.

Why

The previous README was technically accurate but read like library documentation. A new visitor had to understand VES before seeing why it matters. This version makes the trust problem and the demo visible first, then moves into API and implementation details.

Scope

Documentation only. No Core/API/runtime behavior changes.