refactor: streamline CI and improve documentation

[imbajin]

What changes

This PR reduces repeated site-build work, improves the Ask AI entry, and updates both SeaTunnel guides for SeaTunnel 3.0+.

• CI: every PR previously rebuilt all five versions in separate jobs. Verified historical artifacts can now be reused; cold runs use at most three build groups and full warm hits use one. Each runner builds at most two versions concurrently. Version/security validation uses bounded parallelism, and the required deploy gate runs on the aggregate/browser job. Latest is always fresh and every required check remains enabled. Cache misses, corruption and outages fall back to builds; manual candidates remain compatible with older build scripts. setup-node now uses a pinned Node 24 action runtime.
• Ask AI: visiting the site no longer opens the consent notice. Explicit activation prompts for first-use consent; the footer revoke-consent control is removed. Documentation search results scroll independently while the AI action remains visible, including during keyboard navigation. The community link opens HugeGraph Issues. Disabled AI and unconsented use make no third-party AI requests.
• SeaTunnel: EN/CN import and export guides target 3.0+ and follow upstream latest/Next references. Installation examples use a shared SEATUNNEL_VERSION so the released binary and plugins stay aligned. The old 2.x compatibility section is removed. The contribution guide records that the HugeGraph 1.8.0 release snapshot must pin SeaTunnel references and installation to 3.0.0, while latest continues tracking upstream.

Benchmark: core record

All times are seconds. Required path runs from prepare start through the required gate, including inter-job waits but excluding the initial queue. Runner total sums mandatory jobs; advisory visual capture and production publication are excluded.

Sample	Required path	Runner total	Browser contracts	Evidence
Master baseline	603	974	97	Baseline run
Chosen policy, cold cache	551	726	102	Cold run
Chosen policy, warm cache	580	575	97	Warm run

Conclusion: the strongest observed gain is lower runner use, with a modest improvement in elapsed time. Hosted CPU and queue variation matters: these are observed samples, not a guaranteed speedup or a controlled cold-versus-warm comparison. The warm sample predates five additional UI contracts; these measurements also precede the final SeaTunnel content update.

The selected scheduling avoids the single-runner cold regression (697s). Three warm workers gave no build-job improvement over two (141s versus 140s), so the final helper enforces a two-worker limit. Chromium download was about 10s and did not justify another cache.

A controlled local final-security scan of 1,110 HTML files produced identical output and exit status at one/two workers: 28.2s → 15.7s, independently supporting the parsing/concurrency improvement.

Validation

• Python 3.13 source suite: 227 cases, with the rendered-download check executed separately against the complete aggregate. 53 UI unit tests and 14 workflow contracts passed.
• Strict Hugo 0.165 Extended builds of all five versions, aggregate validation, and all 102 Chromium contracts passed at the reviewed implementation head: CI.
• Five independent implementation review lanes completed with no outstanding findings. The SeaTunnel release update received a separate source-grounded review; all 13 unique official documentation/download URLs returned HTTP 200.
• Desktop/mobile before-and-after screenshots and keyboard geometry checks cover the search layout. The illustration summarizes measured results; the table above supplies their source runs and workload differences.
• Latest SeaTunnel/image commit: fresh strict latest build, full artifact validation and all 24 search-ranking contracts passed locally. Latest-head CI validates the updated guides.