chore: split NPM packages

[jif-oai]

Summary

This PR changes npm distribution for @openai/codex from one monolithic tarball to a split-package model:

• @openai/codex becomes a small JS launcher package.
• Native binaries move into per-platform packages (for example @openai/codex-darwin-arm64, @openai/codex-
  linux-x64-musl, etc.).
• The root package uses optionalDependencies so npm installs only the matching platform package.

Why

Our current publish uploads a single large tarball (~206.5 MB, ~504.7 MB unpacked), which now fails in
release CI with:

• npm ERR! code E413
• 413 Payload Too Large - PUT https://registry.npmjs.org/@openai%2fcodex

Splitting by platform keeps each publish payload below registry limits and follows established native-CLI
npm patterns.

Behavior/Compatibility

• npx @openai/codex and codex CLI UX stay the same.
• No runtime binary download is required.
• Unsupported platform/arch still returns a clear error.

Release Workflow Changes

• Build and pack platform-specific npm packages.
• Publish root package + platform packages under the same version/tag (latest / alpha).
• Keep trusted publishing/provenance flow unchanged.

[jif-oai]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6b4e2bdd8d

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".