feat: speed up attw on native packages by ignoring *.node files

[sxlijin]

When running attw on a native package - i.e. one that implements the library logic in C++ or Rust or something via N-API - attw today will read the entire library-aarch64-apple-darwin.node binary artifact, which can easily be 100MiB+, but do nothing with it because there's no type information in the binary artifact.

We ran into this in
BoundaryML/baml#4481 and are working around this by invoking attw through a wrapper that ignores dist/*.node, but this felt like a good thing to upstream :)

(Used my coding agent to open the PR, but I've rewritten the PR title and description to be hopefully useful.)

[changeset-bot]

🦋 Changeset detected

Latest commit: 8318a9e

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 2 packages

Name	Type
@arethetypeswrong/core	Patch
@arethetypeswrong/cli	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[sxlijin]

Original agent PR description:

Summary

• exclude native addon files ending in .node from the core package filesystem
• add regression coverage showing the native binary is ignored while package metadata, JavaScript, declarations, similarly named files, and analysis results are unchanged
• add a patch Changeset for @arethetypeswrong/core

Motivation

BoundaryML/baml#4481 is the motivating real-world example. BAML's ATTW CI currently stages a package without a very large opaque native .node binary because ATTW checks the JavaScript/declaration surface and the native binary is shipped separately in per-platform packages. Filtering *.node upstream would make that staging workaround unnecessary or simpler.

This link is context only; this PR does not close BoundaryML/baml#4481.

Validation

• pnpm format
• pnpm exec prettier --check packages/core/src/createPackage.ts packages/core/test/createPackage.test.ts .changeset/ignore-native-addons.md
• pnpm tsgo
• pnpm build
• FORCE_HYPERLINK=0 TERM_PROGRAM= CI=1 pnpm test (59 core tests and 30 CLI tests)
• pnpm check-dts

The repository does not define a lint script.

[andrewbranch]

Thank you! Where exactly is the work being saved here? Something still reads the Uint8Array from the tarball and passes it to the Package constructor; I’m assuming the actual savings come from something else triggering tryReadFile on quickly returning undefined rather than decoding to string. I would be interested to know what’s trying to do tryReadFile, because the TS compiler itself shouldn’t ever be reading those contents.

[sxlijin]

I'm not totally sure, unfortunately: we had an external contributor send in the attw wrapper improvement to us, which dropped our attw step from 60s to 5s, which seems pretty plausible time savings for cutting out a 630MiB file read.

I will admit that I'm not sure if the implementation that codex put together is the correct one, and I don't actually know what the expensive io op is that this change saves, but this repo seemed low activity enough that upstreaming the issue by opening the PR instead of a GH issue. If I'm reading your comment correctly, are you suggesting that the correct fix should be somewhere else?

[andrewbranch]

Yeah, that’s what I’m guessing, but it’s been a long time since I’ve been in the weeds of this code. I will take a closer look.

[sxlijin]

Sounds good, thanks! Would love if we could drop our wrapper script and just use attw directly :)