chore: consolidate repo structure; delete stale apps/backend fork

[Rakesh1002]

Summary

Collapses the dual-backend layout ADR-0006 had been promising to fix. apps/backend/ was a stale fork of root app/ with diverged code: pre-rebrand UnQuestRequest class names, different import paths (app.services.searxng vs app.services.core.searxng), 2 Alembic migrations vs 5 at root, and (critically) a 2,841-line production SearXNG settings.yml that was never mounted — the live mount path searxng/settings.yml was a 40-line stub falling back to upstream defaults. That's a real bug this PR fixes by salvaging the production config to the live mount path before deleting the fork.

All live deployment paths (ci-cd.yml, deploy-cf.yml, Dockerfile.cloudflare, docker-compose.yml, Makefile, ecosystem.config.js, scripts/manage.sh) already targeted root app/. apps/backend/ existed only as documentation debt promised by ADR-0006 to be collapsed in a future PR.

What changed (134 files)

Phase 1 — Salvage

• apps/backend/searxng/settings.yml (2,841 lines, production-tuned) → searxng/settings.yml. Replaces the 40-line stub at the live mount path.

Phase 2 — Delete

• git rm -r apps/backend/ — 130+ files of stale code (Procfile, Dockerfile, alembic, scripts, tests, etc.).
• Delete .github/workflows/deploy.yml — Railway deploy via cd apps/backend sunset by ADR-0010 (CF Containers GA). Active deploy is deploy-cf.yml.
• Delete scripts/verify-setup.sh — described a Turborepo monorepo this repo never was (packages/shared/, vercel.json, apps/web/src/...). Would have failed 80% of its own checks today.
• Delete setup_webhook_quick.sh — orphaned dev artifact, no references in manage.sh or any deployment script.
• Delete apps/web/public/{file,globe,next,vercel,window}.svg — unreferenced Next.js scaffold SVGs.

Phase 3 — Healthy housekeeping

• Add pnpm-workspace.yaml — SDK packages reference each other via workspace:* but workspace was never declared at root.
• Add apps/sdk-ts/tests/client.test.ts — first vitest test (vitest was configured but no test files existed). Smoke-tests constructor + search() header construction.

Docs reconciled

• docs/adr/0006-monorepo-with-apps-and-workers.md — Amendment section documenting the consolidation; resolved the "Two backend layouts" Con line.
• docs/architecture.md — "Backend Container — app/ and apps/backend/" → app/ only; removed the technical-debt row about dual backend.
• docs/configuration/env-variables.md — apps/backend/.env → .env.

Risk

• Production deploy: unaffected. deploy-cf.yml and ci-cd.yml both target root app/ already. The deleted deploy.yml had not been the active path since CF Containers GA in April 2026.
• CI: .github/workflows/deploy.yml was the only workflow targeting apps/backend. Removed in this PR.
• Self-host: SearXNG behavior improves (real production config replaces stub).
• Working tree: apps/backend/.env (gitignored, contains secrets) is left in the working tree. Manually compare with root .env and remove if redundant.

Test plan

• git pull origin main && ls apps/ shows no backend/ (after manual removal of the gitignored .env).
• wc -l searxng/settings.yml returns 2841.
• pnpm install at root succeeds with the new pnpm-workspace.yaml.
• cd apps/sdk-ts && pnpm test passes the new smoke test.
• CI: ci-cd.yml quality check + deploy-cf.yml worker + container jobs run green on this PR.
• docker compose up -d searxng mounts the 2,841-line config (verify via docker logs unsearch-searxng | head — should not show "using default settings" warning).

Follow-ups (not in this PR)

• Extract shared types between apps/sdk-ts/ and workers/src/lib/ (e.g. VectorMatch) to a packages/sdk-types/ workspace.
• Add .github/workflows/sdk-ts.yml so the new vitest test runs in CI.
• Expand backend test coverage from ~40% to 80% per CONTRIBUTING.md.
• Replace apps/web/public/ Next.js scaffold with UnSearch branding (favicon, OG image, logo).