Fix RolePatch SEO and agent-readable GEO coverage

[sarthakagrawal927]

What

• defines one 20-page public surface registry shared by the sitemap and /api/ai
• serves truthful Markdown through content negotiation and explicit .md routes before OpenNext
• adds public jobs, FAQ, and changelog pages to the canonical sitemap
• fixes missing canonicals and heading semantics on jobs, evidence, and FAQ
• brings over-limit tool and article metadata into the Fleet SEO ranges
• preserves the independently generated Fleet discovery files while keeping curated RolePatch coverage in a separate handler

Why

The live GEO audit could read only 3 of 20 sitemap routes, and only 1 of 3 catalog surfaces had a valid Markdown target. The live SEO audit also found missing canonicals on /jobs and /evidence, no H1 on /evidence, and a skipped FAQ heading level.

Impact

Every cataloged public page now has a same-origin Markdown target, every catalog URL comes from the same list as the sitemap, and private/user-specific application routes remain outside the public corpus. The rendered SEO corrections do not redesign any page.

Root cause

RolePatch's generated edge payload only defined homepage Markdown and left two catalog md fields empty. The sitemap and catalog were maintained separately, so the public jobs, FAQ, and changelog routes could drift out of discovery.

Checks

• pnpm exec vitest run src/__tests__/agent-edge.test.ts src/__tests__/achievement-evidence-bank.test.tsx — 8 passed
• pnpm typecheck
• focused Biome check over all 21 changed files
• pnpm cf:build — Next, deploy-route verification, OpenNext, Astro, and overlay passed
• wrangler deploy --dry-run — final outer Worker bundle contains the curated handler
• local Fleet SEO audits — FAQ, six tools, and two corrected blog articles had zero critical failures
• git diff --cached --check

Closes #30

No production deployment was performed.