perf: share db, docker and auth singletons across duplicated bundles

The @dokploy/server modules get bundled multiple times per process
(esbuild inline copy, compiled node_modules copy, and several Next.js
chunks via transpilePackages), so every module-level side effect ran
once per copy: up to 6 postgres pools, 6 dockerode clients and 6
better-auth instances in the server process, plus the repeated
'Using Docker socket' logs at boot.

- db/index.ts: use the globalThis cache in production too (one pool per process)
- constants/index.ts: cache the dockerode client on globalThis
- lib/auth.ts: wrap betterAuth() in a factory and cache the instance
- Dockerfile: exec node directly instead of leaving pnpm resident (~100MB RSS)

Dockerfile

@@ -69,4 +69,5 @@ EXPOSE 3000
 HEALTHCHECK --interval=30s --timeout=5s --start-period=60s --retries=5 \
   CMD curl -fs http://localhost:3000/api/trpc/settings.health || exit 1
 
-  CMD ["sh", "-c", "pnpm run wait-for-postgres && exec pnpm start"]
+# Ejecutar node directamente: pnpm como wrapper queda residente (~100MB RSS)
+  CMD ["sh", "-c", "node -r dotenv/config dist/wait-for-postgres.mjs && node -r dotenv/config dist/migration.mjs && exec node -r dotenv/config dist/server.mjs"]