Withdraw conformance and benchmark claims sitewide; decouple site build from wasm

[MelbourneDeveloper]

TLDR

Withdraws Basilisk's 100% conformance claim and its published benchmark figures across the whole public site (English and Chinese), replaces them with a linked correction and the integrity audit that explains why, and decouples the Eleventy build from the Rust wasm build so a non-compiling checker can no longer take the entire site down.

What Was Added?

• docs/CONFORMANCE-INTEGRITY-AUDIT.md (285 lines) — the record of what was found: checker logic fitted to the exact contents of conformance fixtures rather than implementing the typing specification generally, including a type-alias validator matching prefixes and substrings of raw source text with a special case for eval( because that spelling appeared in one test file.
• Explicit Build playground engine (wasm) step in both ci.yml's website job and deploy-pages.yml, replacing the implicit one that npm run build used to chain.
• .bench-caveat notice on /docs/benchmarks/, stating the figures are withdrawn pending an integrity review and must not be used to compare Basilisk with other tools.
• New e2e assertions that the old claims are absent: expect(body).not.toContain("scores 100%") and expect(body).not.toContain("fastest we've benchmarked").

What Was Changed or Deleted?

• Homepage (EN + ZH) — headline drops "the only Python type checker that scores 100% … and the fastest we've benchmarked" for "An open-source Python type checker and language server, built in Rust." The two hero proof cards now read Current typing conformance and Published benchmark figures, both pointing at the withdrawal.
• /docs/conformance/ — retitled "Basilisk Conformance Results Are Withdrawn"; the leaderboard table, the "How it's measured" section and the reproduce-it-yourself instructions are removed. States plainly that Basilisk was removed from the python/typing results table at our request and that its current percentage is temporarily unknown.
• _data/conformanceOfficial.js — no longer exports live claim data (basiliskRank, perfectCount, basiliskIsSolePerfect). It now exports withdrawn: true, publicationStatus: "historical-withdrawn", and nests the frozen snapshot under historical.* with basiliskWasSolePerfectAtSnapshot. resultsDir is repointed from main to the pinned commit so the snapshot cannot silently re-render as current.
• website/package.json — "build" changes from npm run build:wasm && eleventy to eleventy. Every page but the playground renders from committed data, so the site build no longer has a Rust dependency.
• Site/social metadata — eleventy.config.js description, package.json description, site.json, og-image, banner SVG/PNGs and the og:title all updated; page title drops "Fast".
• Blog posts (EN + ZH) — the 100%-conformance announcement and five other posts updated so no retracted claim survives in an older article.
• /docs/comparison.md, /docs/index.md, quick-start, migration, install-vscode, releases.njk, benchmark-section.njk, conformance-chart.njk — comparative claims removed or re-scoped.

How Do The Automated Tests Prove It Works?

npm run test:e2e — 52 passed, 6 skipped, 0 failed on both the desktop (Desktop Chrome) and mobile (iPhone SE, 375×667) projects, run with CI=1 so reuseExistingServer is off exactly as in CI.

The suite proves the correction is actually rendered, not just committed:

• homepage.spec.ts › puts the integrity correction beside the product introduction — asserts the subheadline contains "withdrawn both our former conformance claim and our published benchmark figures", "removed from the official python/typing results at our request", "current conformance percentage is temporarily unknown", and "robustness and mutation testing".
• homepage.spec.ts › shows both withdrawn result notices and their detail links — asserts exactly 2 stat cards, "removed from the official results table", "Both sets of figures are withdrawn", and that the body no longer contains "scores 100%" or "fastest we've benchmarked". This is the assertion that fails if a retracted claim is reintroduced anywhere on the page.
• homepage.spec.ts › Chinese homepage › renders the same correction and withdrawn-result notices — the same gates in zh, so one locale cannot quietly keep a claim the other withdrew, plus keeps its disclosure structure aligned with the English page.
• benchmarks.spec.ts — asserts .bench-caveat contains "benchmark figures are withdrawn pending an integrity review" and "Do not use the values below to compare Basilisk with other tools".
• playground.spec.ts (4 tests × 2 viewports) — drives the real wasm engine, proving the npm run build decoupling did not break the playground.

Also run locally and green: python3 scripts/gen_rules_reference.py --data + diff (rules.json in sync, 169 codes), python3 scripts/gen_conformance_reference.py --check (stamped references and generated READMEs in sync), npm run build:wasm under CI's RUSTFLAGS="-D warnings -C link-arg=-zstack-size=16777216", npm run build (302 files), cargo fmt --all --check (0 diffs) and cargo check --workspace --all-targets.

make ci (lint test build) was not run in full: this branch changes no Rust, and CI path-filters those jobs off for it (code=false). cargo check --workspace --all-targets and cargo fmt --all --check both pass.

Spec / Doc Changes

• docs/CONFORMANCE-INTEGRITY-AUDIT.md — new; linked from the site's conformance correction.
• docs/INDEX.md — registers the audit under a new Conformance integrity section.
• docs/specs/WASM-SPEC.md — records that the playground engine is built by an explicit build:wasm step, separate from the Eleventy build.
• docs/specs/WEBSITE-E2E-SPEC.md — [WEBSITE-E2E-SMOKE] rewritten: the homepage gates now assert the integrity disclosures and the absence of the withdrawn claims, in both locales.

Breaking Changes

• None to any shipped binary or API. _data/conformanceOfficial.js changes shape (live fields replaced by withdrawn + historical.*), which is internal to the site build and consumed only by templates updated in this PR.

Refs #379, #408