Fix domain surface consistency across landing page and registry

[rksharma-owg]

Summary

• fix the root cause where domain-level surface formats were derived from different data sources on the homepage, baseline domain pages, and registry summaries
• add shared logic to combine curated provider interfaces, discovered-domain catalog surfaces, and normalized catalog records into one ordered, deduplicated set of domain surface types
• reuse that shared logic for homepage cards, prerendered domain pages, baseline /disc/{domain}.json, and surface sitemap generation

Root cause

The landing page and registry summary paths were using normalized catalog/index data, while prerendered domain pages and baseline discovery JSON were deriving visible surface types from a narrower baseline that could miss discovered-domain-only formats like CLI. That let a domain such as Notion show CLI in the registry inputs but not on the landing-page/domain representation.

Shared logic introduced

• src/lib/domain-surface-types.ts normalizes surface type vocabulary to the project's existing mcp, openapi, graphql, cli ordering and deduplicates them
• src/lib/domain-discovery.ts builds a consistent domain discovery document by combining discovered-domain surfaces, curated interfaces, and baseline catalog surfaces
• src/lib/curated.ts, src/lib/discovered-catalog.ts, and src/lib/discovered-domains.ts provide reusable loaders/mappers so domain-level surface joins happen in one place rather than ad hoc in page components

Pages/components changed

• homepage / registry summary path via src/lib/catalog.ts
• prerendered domain page via src/pages/[domain].astro
• baseline domain discovery JSON via src/pages/disc/[domain].json.ts
• surface sitemap generation via src/pages/sitemap-surfaces.xml.ts

Regression tests added

• Notion includes CLI when the underlying catalog contains it
• duplicate surface types appear only once
• curated-only surfaces remain visible
• raw-catalog-only surfaces remain visible
• a domain with no surfaces is handled safely

Commands run

• bun install ✅
• bun run normalize ✅
• cd cli && bun run build ✅
• bun test ✅ (72 pass, 0 fail)
• bun run build ✅
• bun run validate:batch ⚠️ fails due to pre-existing catalog data issues unrelated to this fix (catalogMissingLocator, catalogDuplicateSurface)

Before / after for Notion

Before:

• the domain/landing-page representation for Notion omitted CLI
• the underlying discovered/catalog data could still include a Notion CLI surface

After:

• homepage/domain summary includes CLI for Notion
• prerendered notion.com domain output includes MCP · REST · CLI
• baseline /disc/notion.com.json includes the Notion CLI surface
• /api/domains.json for notion.com includes cli in formats

Fixes #41