[Mock API] Experiment: mock endpoint on the website via Cloudflare (ADR Option 3B)

[karinamzalez]

Summary

Summary

Build a throwaway, high-fidelity experiment of the integrated endpoint shape from the mock playground ADR so we can measure the actual surface area of the change before committing to a direction.

In this shape, the mock API is a server-rendered endpoint on the website itself, with the site deploying to Cloudflare instead of GitHub Pages. Same origin as the docs, one deployment pipeline, and co-located with the specs, schemas, and sample-data generation the site already has.

This is the companion to the Option 3A experiment and uses the same rubric. The hypothesis to test: because the specs, schemas, and sample-data generation already live in the website, the surface area of this change may be smaller than "migrate hosting" makes it sound and it avoids keeping a second service in sync.

Data layer: same approach as the 3A experiment-- start from the website's existing sample-data generator (generateSchemaExample() in src/lib/schema/example-generator.ts, which runs openapi-sampler over the JSON schemas and powers the examples on the models pages), falling back to the #1049 fixture/handlers only where it can't deliver. Note the generator reads files at build time, so even here samples likely need pre-generating rather than running per-request-- confirm.

Note this work is likely not wasted even if we choose 3A: the DNS migration to Cloudflare is happening regardless, and parallel GH Pages + Cloudflare deploys (e.g. beta.commongrants.org) are a useful stepping stone. If we don't choose this direction, it's a PR we close.

Acceptance criteria

Acceptance criteria

• Astro configured for server output with the Cloudflare adapter, deployed to a Cloudflare preview, with existing static docs pages still rendering correctly
• Mock opportunity endpoints served from the website itself, same origin as the docs (no CORS needed)
• Swagger UI "Try it out" successfully calls the endpoint
• Continuity check: the same record fetched from the browser, from a copied curl command, and from the TS SDK returns identical data
• Findings written up against the shared rubric:
  • Surface area: what files, configs, and pipelines changed (rough LOC + list); what breaks in the static build
  • Data layer: how far the existing sample generator gets us against the [POC] #1034 Test A: client-side MSW mock playground #1049 review criteria-- deterministic, list↔detail consistent, semantically representative, filterable/sortable, error responses-- and what thin layer is needed on top for the gaps
  • Sync story: whether co-location makes staying in sync meaningfully easier than in the standalone shape, or whether both need the same build-time pre-generation
  • Deployment overhead: what changes about the publish pipeline; what the GH Pages → Cloudflare cutover would involve
  • Risks and unknowns discovered
• Rough effort estimate for productionizing this shape
• Timeboxed, if it runs long, stop and write up what's known