perf(search): load Pagefind lazily on first open; build the index in dev and build

[Vidminas]

🚀 What type of change is this?

• 🐛 Bug fix (A non-breaking change that fixes an issue)
• ✨ New feature (A non-breaking change that adds functionality)
• 💅 Style change (A change that only affects formatting, visuals, or styling)
• 📚 Documentation update (Changes to documentation only)
• 🧹 Refactor or chore (A code change that neither fixes a bug nor adds a feature)
• 💥 Breaking change (A fix or feature that would cause existing functionality to not work as expected)

---

🎯 What is the purpose of this change?

The search modal imported Pagefind eagerly in init(), so pagefind.js and the index were
fetched on every page view, even for visitors who never open search. (The current try/catch +
assetPaths already make a missing index degrade gracefully — this is the remaining perf/ergonomics
refinement.)

Changes:

1. Lazy load (engine). Move the import into an idempotent ensurePagefind() that runs the first
   time the search modal opens (via the existing $store.search.open watcher), and is awaited by
   search() so typing before the load finishes still works. Net: the index is fetched on first
   search-open, not on page load.

2. Build the index in dev and build (all starters). Inline the Pagefind call in every
   starter's dev and build scripts, replacing the pnpm run pagefind indirection so each script
   is explicit and self-contained:

   • dev: hugo && pagefind --site public --output-subdir ../static/pagefind && hugo server …
     — builds the index into the already-gitignored static/pagefind/ so hugo server serves it,
     making search work under pnpm dev (mirrors the monorepo dev harness in view-starter-dev.sh).
   • build: hugo --minify && pagefind --site public — writes the index straight into the
     deployed public/pagefind (the standard location; unchanged behaviour, just inlined).

   The standalone pagefind script is kept because each starter's netlify.toml still calls
   pnpm run pagefind. The starter base template wraps its 2>/dev/null || true fault-guard in a subshell —
   hugo && (pagefind … || true) && hugo server — so it tolerates only a Pagefind failure. Inlined
   bare, sh's equal &&/|| precedence would let || true also swallow a failed Hugo build and
   start the server on broken output.

Verified: the built search script has exactly one await import(...) (inside ensurePagefind),
init() no longer imports eagerly, the open-watcher calls ensurePagefind(), and the emitted JS
passes node --check. All nine starters' package.json remain valid JSON, and Pagefind's output
locations are confirmed empirically — --output-subdir ../static/pagefind writes to
static/pagefind/ (dev, served by hugo server) and the default --site public writes to
public/pagefind/ (build, the deployed output).

---

📸 Screenshots or Screencast (if applicable)

n/a

---

ℹ️ Documentation Check

• No, this change does not require a documentation update.
• Yes, I have updated the documentation accordingly (or will in a follow-up PR).

---

📜 Contributor Agreement

Thank you for your contribution!

• By checking this box, I confirm that I have read and agree to the HugoBlox Contributor License Agreement (CLA).

[netlify]

❌ Deploy Preview for academic-demo failed. Why did it fail? →

Name	Link
🔨 Latest commit	d04d53a
🔍 Latest deploy log	https://app.netlify.com/projects/academic-demo/deploys/6a535e1744e8130008fa89d0

[Vidminas]

Builds fail due to Hugo 0.162.0 incompatibility, which is addressed in #3329