perf(data-pages): replace O(N) linear-scan with direct /v1/<kind>/:slug lookup

[chrisjmendez]

Closes #9 (partially — see scope note).

What changed

Two data profile pages were fetching 500 records and then .find()-ing the slug we actually needed. The REST API already exposes GET /v1/<kind>/:slug (mounted by buildDomainRouter), which does the lookup in O(1). Switching cuts per-page bytes-over-the-wire by ~100× and removes a latent CF-throttle risk during ISR revalidation.

Page	Before	After
apps/web/src/app/data/compatibility/[slug]/page.tsx	fetch(/v1/compatibility-edges?limit=500) + .find()	fetch(/v1/compatibility-edges/${slug})
apps/web/src/app/data/deploys/[slug]/page.tsx	new FestivusClient().searchDeployNotes({ limit: 500 }) + .find()	fetch(/v1/deploy-notes/${slug})

Both now mirror the pattern already used by apps/web/src/app/data/benchmarks/[slug]/page.tsx.

Scope note: papers excluded

Issue #9 lists three pages. This PR refactors two. The third — apps/web/src/app/data/papers/[slug]/page.tsx — needs separate API work first: GET /v1/papers/paper16 returns 404 not_found today even though the Prisma papers table has a slug column. The seeded slug values do not match the in-JSON id field that the URL passes through. Fixing that is a schema/seed alignment, not a page refactor; it gets its own PR.

Verify

• pnpm check (typecheck + lint): ✅ green
• pnpm test: 605/606 — the one failure is the local-api canary that needs a running localhost:8000. Re-run against the production API: FESTIVUS_DATASET_API_URL=https://api.festivus.hapticlabs.ai npx vitest run tests/seed-validation/festivus-api-canary.test.ts → 1/1 passes
• Live API spot-check before publish:
  • curl -o /dev/null -w "%{http_code}" https://api.festivus.hapticlabs.ai/v1/compatibility-edges/franka-research-3__openvla-openvla-7b → 200
  • curl -o /dev/null -w "%{http_code}" https://api.festivus.hapticlabs.ai/v1/deploy-notes/deploy-arm-1 → 200

DCO

Signed-off-by trailer on the commit.

Test plan after merge

1. Hit any compat-edge profile page in prod (e.g. /data/compatibility/franka-research-3__openvla-openvla-7b) → still renders.
2. Hit any deploy-note profile page → still renders.
3. Network tab confirms each page fires one request, not a limit=500 list.

Refs: #9