refactor: simplify Effect server architecture

[ScriptedAlchemy]

Summary

• Create one scoped Effect ManagedRuntime for each plugin setup.
• Let that runtime own route watchers, delayed lazy-compilation prewarm work,
  type generation, prerender workflows, background fibers, and finalizers.
• Dispose resources in a deterministic, idempotent order from every Rsbuild
  shutdown path.
• Keep PR perf: keep route transforms inline by default #86's newer inline/opt-in route-transform architecture; the removed
  bespoke transform worker is not reintroduced.
• Update Effect to 3.22.0 and the Rstest 0.11.x range to at least 0.11.4.

Current-main repair

This branch was rebuilt from current main after #74 and #86 merged. It is now
a single commit and contains none of the obsolete pre-squash #74 history.
Conflict resolution preserves the verified RR8/RSC fixes from #74 and the
transform-loader design from #86.

Maintainability

• One runtime creation site and one ordered shutdown policy.
• Effect Scope, FiberSet, and ManagedRuntime replace manual watcher,
  delayed-task, and cleanup coordination.
• Production src/ is 100 lines smaller than current main (631 additions,
  731 deletions).
• The 1,232-line implementation plan was removed from the shipping diff.

Release

• Adds a patch changeset describing the internal lifecycle improvement.
• Existing changesets still determine the combined next package version.

Validation

• pnpm test — 61 files and 639 tests passed, including all TypeScript
  project checks.
• pnpm build.
• pnpm test:package-interop.
• pnpm e2e — all selected classic SSR, SPA, prerender, custom-server,
  Cloudflare, React Router 8, and RSC development/production examples passed.
• pnpm format:check, pnpm changeset status, and git diff --check.
• Emitted transform loaders and browser/RSC runtime templates are explicitly
  checked to contain no Effect imports.

Performance

This is a lifecycle/maintainability change, not a claimed hot-path speedup.
Effect remains outside emitted transform loaders and browser/runtime templates;
the current-head benchmark checks are left to CI rather than carrying forward
stale measurements from the old stacked branch.

[pkg-pr-new]

Open in StackBlitz

pnpm add https://pkg.pr.new/rsbuild-plugin-react-router@779d017

commit: 779d017

[github-actions]

Benchmark results

Case	Base	Head	Delta
build-256-ssr	1994.3 ms	2001.2 ms	+0.3%
dev-48-ssr	865.9 ms	852.1 ms	-1.6%

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

✅ 2 untouched benchmarks

---

_{Comparing codex/effect-server-architecture (779d017) with main (0dd1fa8)}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 779d017fee

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Run every merged buildEnd hook before rejecting

When multiple presets and/or the user config provide buildEnd, a rejection from configA.buildEnd now prevents configB.buildEnd from being invoked. The previous Promise.all/Effect.all composition started every merged hook, so this can skip required cleanup or artifact-finalization work whenever an earlier hook fails; preserve deterministic ordering if desired, but still attempt all hooks before propagating their failures.

Useful? React with 👍 / 👎.

[ScriptedAlchemy]

Addressed in #109 with a regression test that fails when a later merged buildEnd hook is skipped. The follow-up preserves deterministic order, runs remaining hooks, and rethrows the first normalized failure.