unlock dev setup, add service healthchecks, fix compose mounts

[Rajaykumar12]

Summary

Addresses three findings from the hosting readiness audit: a fresh clone couldn't start
the dev stack, server/client had no restart policy or healthcheck in production, and
APP_ROOT silently installed to the wrong place.

Changes

Dev setup

• Add server/.env.example and client/.env.example. client/.env in particular needs
  BACKEND_URL=http://localhost:8080 — the built-in default http://server:8080 is a
  Docker-network name that doesn't resolve on the host.
• Rewrite CONTRIBUTING.md Step 1–2: explain what each of the three env files is for, and
  replace the stale compose command (which referenced the deleted seed service) with
  pnpm install && pnpm dev.

Health and resilience

• New GET /health on the API, returning 503 until Mongo is connected.
• restart: unless-stopped on server and client, matching mongo and caddy.
• Healthchecks on server and client; caddy and client now gate on
  condition: service_healthy instead of start-order, so Caddy can no longer proxy to a
  Next.js that isn't listening yet.

APP_ROOT

• Prod compose bind mounts and env_file paths resolve ${POMELO_ROOT:-/opt/pomelo}
  instead of hard-coding /opt/pomelo. Default behaviour is unchanged.
• ComposeCommand passes POMELO_ROOT: paths.root to compose; install.sh does the same
  on uninstall teardown, and exports NODE_ENV=production in the systemd-less fallback so
  a relocated install doesn't trip the dev-mode heuristic.