perf: box large compilation artifacts

[intellild]

What changed

This PR boxes several large compilation artifacts across async and persistent-cache boundaries so they are no longer returned or moved by value through hot paths. The main flow now carries boxed build-module-graph artifacts, exports-info artifacts, build results, parse results, code-generation results, and persistent-cache artifacts.

It also simplifies the large BindingCell constructors back to new(Box<T>), adds From<Box<T>> support where needed, and updates affected module/plugin implementations to keep large results boxed through the call chain.

A new rspack_large_stack_type dylint crate was added to catch known large Rspack types when they appear in stack-sized positions. The root lint config also enables relevant clippy/rust lints for large assignments, futures, and stack frames.

Why

These artifacts can become large enough that returning or moving them by value increases stack pressure and unnecessary data movement. Boxing them at ownership boundaries keeps async futures and compilation-stage handoffs smaller while preserving the existing ownership model.

Expected impact

The change should reduce stack usage and large moves in build-module-graph, parser/build/codegen, and persistent-cache recovery paths. It also gives future changes a guardrail so new large artifacts are passed as Box or references instead of silently growing stack frames.

Checked locally with cargo fmt --all --check, focused cargo check for core/devtool/swc-minimizer, cargo test --locked --no-run for the new dylint crate, cargo check -p rspack_benchmark --benches, and cargo check --profile codspeed -p rspack_benchmark --bench benches --bench rspack_sources. The dylint UI run reached the temporary driver stage locally, but the driver process did not complete reliably in this environment.

[github-actions]

📦 Binary Size-limit

Comparing ac8b64a to fix(esm_library): preserve static source-phase external bindings (#14674) by Elecmonkey

🎉 Size decreased by 16.00KB from 66.74MB to 66.73MB (⬇️0.02%)

[codspeed-hq]

Merging this PR will not alter performance

✅ 43 untouched benchmarks
⏩ 47 skipped benchmarks¹

---

_{Comparing perf/box-large-artifacts (ac8b64a) with main (df598e4)}

Footnotes

1. 47 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩