Rewrite the README top around evidence a reviewer can verify

[weiminglong]

Summary

The front page previously led with a generic data-collector description and contributor boilerplate; the repo's strongest evidence sat buried in docs/. The top now delivers, in order:

1. A systems pitch naming the actual mechanisms (lock-free single-writer ingestion, zero-copy WS deserialization, CRC-framed WAL with torn-tail recovery and flock failover, deterministic reconstruction).
2. The live-demo link and a screenshot of the workstation streaming the demo market (docs/assets/workstation-demo.png, captured from a production build — no devtools chrome).
3. A measured hot-path medians table (from the committed PERFORMANCE.md run) — the first thing a quant reviewer looks for.
4. A "Review this repo in 15 minutes" tour routing to the WAL internals, TESTING.md's failure-mode matrix, the performance report, the research notebook, and the eleven ADRs.
5. A trimmed Project Status.

Near the tail, a new Engineering process section states plainly how the project is built — AI-assisted under human design direction, verified by hard gates (fuzzing, property tests, miri, golden fixtures, offline incident simulations, adversarial review, live end-to-end runs) rather than review-only trust — and where design decisions are recorded. The Benchmarks section no longer repeats the headline numbers now that the top table carries them.

Verification

• Screenshot captured via Playwright against a vite build production preview in demo mode (live simulator ticking, sequence advancing, no dev overlays)
• All links resolve against current main (TESTING.md, docs/PERFORMANCE.md, research/, docs/adr/); hygiene gate clean; doc + asset only