chore: align docs to reality, add build to CI, fix the broken prod build

[MANVENDRA-github]

The repo advertised more than the code shipped, and the gateway production
build was silently broken (CI never ran it). This is the first slice of a
planned series that closes the gap between the docs and the code.

What changed

• Fix: tsconfig.build.json no longer globs the dashboard's .ts sources
  into the Node build (they failed on extensionless imports — pnpm build
  exited 2 on main). Build is green again, and pnpm build is now part of
  the CI verify job so it can't silently regress.
• Reconcile package.json versions 0.0.0 → 0.1.0 (root + both workspaces)
  to match the published release; drop the stale "Scaffold" gateway description.
• Doc-accuracy pass: k6 → Node load harness (load/run.ts); v1.0 → v0.1.0;
  single-node in-memory cache + SQLite (Redis/Postgres described as documented
  future swaps behind existing interfaces); dashboard marked shipped; providers
  described as OpenAI-compatible (native Anthropic noted as planned); honest
  benchmark caveats; new README "Known limitations" section.

Why

• A post-v0.1.0 audit found doc/code drift and a red-but-unrun production build.

How to verify

• pnpm verify → green (typecheck + lint + 90% coverage)
• pnpm build → green (was exit 2 on main)
• npx prettier --check CHANGELOG.md → clean

No runtime behavior change.

Assumptions

• Redis & Postgres are downscoped to docs (v1 is single-node per PRP_SPEC §3/§6),
  not built. Cost tracking and a native Anthropic adapter land in follow-up PRs.