Give every route its own title, description and link preview

[laveshparyani]

The build is a single page, so every URL shipped one head. Google saw five
pages wearing the same name, and a link shared on WhatsApp previewed as the
homepage whichever page was actually sent.

src/data/seo.js is now one table with two consumers.

Build time — vite.config.js

Emits a real HTML file per route, each with its own title, description,
canonical and Open Graph tags. Nothing is rendered server-side: the body is the
same empty root div and React boots exactly as before. Only the handful of tags
a crawler or link scraper reads before it stops are different — and those were
the ones that were wrong.

This matters because link scrapers never run JavaScript. WhatsApp, Facebook
and Twitter read the shipped HTML and stop, so a client-side-only fix would have
left previews broken while looking correct in the browser.

Two files per route (services.html and services/index.html) because which
form GitHub Pages prefers is not worth guessing at when both cost 4kB. They
share a canonical, so the duplicate collapses to one URL.

Runtime — src/lib/seo.js

Updates the live document on client-side navigation by mutating the tags the
build already shipped. React 19 can hoist <title> and <meta> on its own, but
it appends rather than replaces, so rendering them would leave two titles and
two descriptions on every page — and the browser keeps the first.

Also

• /projects and /resources route and render but are still placeholders. They
  now carry noindex and are out of the sitemap. They were previously
  advertised to Google at priority 0.9 and 0.7.
• Unknown paths drop the canonical rather than keep the previous page's. A
  stale canonical claims the 404 is a duplicate of a real page, which is a worse
  signal than making none. Verified the tag is re-created on the next real route.
• sitemap.xml and robots.txt are generated from the same table instead of
  maintained by hand, so they cannot drift again. The old one already had.

Verification

• scripts/verify-seo.mjs, wired in as its own CI step via npm run verify:seo.
• Confirmed the guard fails on a seeded regression, not just that it passes:
  reverting one page's title to the shell title trips both the title check and
  the shared-title check.
• Walked the router through home → services → projects → unknown → about →
  unknown → home, asserting tag counts at every step. Exactly one title, one
  description and one og:description throughout; canonical present on real
  routes, absent on unknown ones, re-created afterwards; robots added and
  removed correctly.
• Existing CI guards still pass.