feat: add local docker compose development stack

[Fabluchy]

Closes #125

Summary

Adds a reproducible local Docker Compose development stack — API, wallet-provisioning worker, PostgreSQL, and Redis — started with one command. While building it, I found and fixed several things that blocked local development entirely (fresh-install Prisma failures, a crash on boot under real Node ESM, and a migration that always fails on a fresh database).

Changes

Docker Compose stack (#125)

• docker-compose.yml — api, worker, db (Postgres 16), redis services
  • Health checks on every service (pg_isready, redis-cli ping, /health/live) with depends_on: service_healthy so readiness waits for dependencies
  • API entrypoint applies migrations and seeds deterministic fixtures on boot; worker entrypoint runs prisma migrate deploy before starting
  • Named volumes (pgdata, redisdata) — reset is project-scoped via docker compose down -v
  • stop_grace_period: 30s matching the app's graceful-shutdown handler
• docker/Dockerfile.dev — dev image: all deps, OpenSSL (for Prisma), generated client; source bind-mounted for nodemon hot reload
• docker/entrypoint-dev-api.sh / docker/entrypoint-dev-worker.sh — explicit migrations + seed, then dev server / worker
• src/workers/wallet-provisioning.worker.ts — polls the idempotent wallet-provisioning outbox (Prisma repo + dev in-memory KMS + SDK keypair generator), with SIGTERM/SIGINT graceful shutdown
• scripts/stack-smoke-test.sh — validates compose config, starts the stack, waits for /health/ready, probes liveness/readiness, prints service status (pnpm stack:smoke; --validate for config-only)
• package.json — stack:up/down/reset/logs/validate/smoke, worker:dev, db:generate, db:deploy
• docs/DEVELOPMENT_STACK.md — quick start, health checks, reset, logs, smoke test, troubleshooting; linked from README
• .env.example — POSTGRES_*, REDIS_PORT, API_PORT, WORKER_POLL_INTERVAL_MS
• CI: validates docker compose config --quiet on every PR

Fixes found while building the stack

• prisma.config.ts — prisma generate failed without a DATABASE_URL (Prisma's env() helper throws), which also meant the production Dockerfile build stage could never generate the client. Now falls back to a local default; seed command moved into the Prisma 7 config (migrations.seed)
• src/config/jwt.ts — import { JsonWebTokenError } from 'jsonwebtoken' crashes the server under real Node ESM (The requested module 'jsonwebtoken' does not provide an export named 'JsonWebTokenError'). Uses the default import instead, matching src/utils/jwt.ts
• Migration ordering — 20260718000000_add_learner_preferences referenced users before 20260719100042_first_migration created it, so prisma migrate deploy always failed on a fresh database (P3009). Renamed to 20260719100043_add_learner_preferences (the old one never applied anywhere — it always failed). 100% rename, no SQL change
• Dockerfile — healthcheck probed /health, which 404s (routes are /health/live + /health/ready); now probes /health/live
• Removed accidentally committed UTF-16 error-log scratch files (server_error*.txt, lint_results*.txt, lint_output.json) — already excluded by .dockerignore

Verification

• pnpm lint — clean
• tsc --noEmit — clean
• pnpm test:ci — 641 passed / 3 skipped (pre-existing DB-requiring skips; no Postgres in unit-test env)
• docker compose config --quiet — valid
• Full smoke test from a clean state (docker compose down -v && scripts/stack-smoke-test.sh):
  • db/redis healthy, api healthy, worker running
  • GET /health/live → {"status":"ok",...}
  • GET /health/ready → {"status":"ready","checks":{"database":"ok"}}
  • Seed deterministic fixtures verified in Postgres (6 users, 8 modules, 16 completions, 14 credentials); re-running is idempotent
  • docker compose stop api → graceful shutdown observed; down -v removes only project volumes