docs: add multi-module guides, benchmarks, and CI benchmark workflow

[deepjoy]

Summary

• Add benchmark suites for retry, dependencies, tags, groups, and history
  (complementing the existing scheduler benchmarks)
• Add CI workflow that tracks benchmark performance over time on a
  _benchmarks branch and posts critcmp comparison comments on PRs
• Add multi-module and library-modules documentation guides; expand
  quick-start, query-apis, priorities, and design docs
• Format bench files consistently

Changes

Benchmarks

• benches/retry.rs — backoff strategies, permanent failure, dead-letter paths
• benches/dependencies.rs — linear chains, fan-in dispatch at varying depths
• benches/tags.rs — submit-with-tags, query/count/scan by tags at scale
• benches/groups.rs — per-group concurrency gates, group scaling
• benches/history.rs — paginated history, aggregate stats, filter-by-type

CI

• .github/workflows/benchmarks.yml — runs on push to main (stores
  baseline via benchmark-action/github-action-benchmark + caches critcmp
  export) and on PRs (restores baseline, compares with critcmp, posts or
  updates a PR comment with the diff)

Docs

• docs/multi-module-apps.md — guide for multi-module scheduler patterns
• docs/library-modules.md — reference for per-module API surface
• Updates to quick-start, query-apis, priorities, design, glossary,
  io-and-backpressure, persistence-and-recovery, migrating-to-0.4