Add detached lifecycle mode for persistent clusters

[m3ridian-zero]

Summary

Refs #28.

Adds an opt-in lifecycle mode for callers that need a persistent embedded Postgres cluster to outlive the process that started it.

• adds lifecycleMode: 'managed' | 'detached'
• keeps managed as the default so existing process-exit cleanup behavior is preserved
• excludes detached instances from the module-level exit hook
• starts detached clusters outside the parent terminal/process lifecycle
  • Windows: CreateProcessW with CREATE_NO_WINDOW, tracking the postmaster PID for explicit stop()
  • POSIX: detached spawn with unref'd process/stdio handles
• waits for detached startup by polling Postgres with SELECT 1
• keeps explicit stop() supported for detached clusters
• retries data-directory removal to avoid transient Windows handle-release races
• documents the new option and adds a detached integration test

Verification

• npx -p node@20 node node_modules/lerna/cli.js run build
• npx -p node@20 npm run lint from the embedded-postgres package
• npx -p node@20 npm test -- tests/index.test.ts -t "detached cluster"
• npx -p node@20 npm test -- tests/index.test.ts -t "should be able to start and stop a cluster"

Note: the full Windows package test file still has an unrelated pre-existing permission-bit assertion for .exe files; I left that behavior unchanged and verified the PR-specific detached path plus the managed default start/stop path.