docs(site): curate the blog around reader questions

[LauraGPT]

Summary

• Replace the bilingual blog's 30-card release/article wall with one lead and four curated stories.
• Add applications, selection and technical-explanation pages; retain all 35 historical article URLs per language in a complete archive and keep release notes separate.
• Re-edit five bilingual stories around a reader question, example and next step. Preserve original dates, code, source links, factual limits and historical evidence.
• Keep the existing static build, shared navigation and search/sitemap; add a validated bilingual catalogue instead of another hand-maintained index.
• Fix near-white FunClip body text caused by a shared CSS variable collision and make inline article links visibly underlined.

Verification

• 476 product-site tests, with exact collected-node coverage across 421/28/27 shards.
• 130 Chromium cases including bilingual mobile/desktop navigation, screenshots, article text contrast and inline-link cues.
• 280 repository documentation contracts (3 existing deselections), 46 Sphinx-link tests and vLLM guide synchronization.
• Two validated, byte-identical builds: 267 files and 202 HTML pages.
• Independent source review; real text-contrast and link-cue failures reproduced before repair.
• Signed/DCO commit, frozen source and verified source/build/evidence rollback backup.

Scope

No new model capability, acoustic benchmark, package release or PyPI publication. Old article URLs remain valid. Legacy blog index snapshots are provenance only; published ordering is owned by data/blog.json and the generated view.