Refactor exports for improved tree-shaking and add sideEffects declarations

[it-rec]

This PR improves tree shaking across Carbon packages by adding explicit sideEffects declarations and removing module-scope patterns that bundlers cannot prove side-effect-free.

Changelog

Changed

• Added sideEffects declarations to package.json:
  • @carbon/layout, @carbon/motion, @carbon/type, @carbon/grid, @carbon/elements: only style entry points (index.scss, scss/, css/) are declared side-effectful
  • @carbon/icons, @carbon/pictograms: "sideEffects": false
• Annotated top-level token initializers (rem(), px(), miniUnits()) in @carbon/layout and @carbon/type with /*#__PURE__*/
• @carbon/type: exported font weights, font families, and type scale steps (scale01–scale23) as scalar constants and referenced them as plain identifiers in token definitions — module-scope member accesses like fontWeights.regular or scale[0] read as potential getter side effects to bundlers and block tree shaking of otherwise-unused tokens
• @carbon/type: fluid() computes the breakpoint name list lazily instead of at module scope
• @carbon/type: inlined the two spread-based expressiveHeading01/02 tokens so each token can be dropped independently

Unchanged

• Public APIs, export lists (verified: 69 exports in @carbon/type, zero diff vs. published), and all token values (deep-compared built outputs against the published packages)
• @carbon/react and @carbon/web-components configuration — @carbon/react already tree-shakes correctly, and web-components' custom-element registration is an intentional side effect

Bundle size impact

Measured by bundling a single token import (e.g. import { blue60 } from '@carbon/elements') in production mode, React externalized, one isolated build per entry. "Before" = published packages (@carbon/type@11.61.0 et al.); "after" = this branch. Sizes are minified bytes (gzip).

Bundler	@carbon/elements	@carbon/type	@carbon/layout
webpack 5	1,351 (456) → 46 (78)	1,420 (505) → 176 (175)	309 (186) → 69 (96)
esbuild	6,140 (1,313) → 32 (64)	6,131 (1,303) → 142 (149)	518 (301) → 63 (89)
Vite 7 (rolldown-vite)	2,221 (756) → 23 (55)	2,278 (755) → 121 (135)	288 (174) → 56 (86)
Rollup 4 (+terser)	24 (56) → 24 (56)	136 (150) → 126 (143)	39 (68) → 47 (76)¹

¹ +8 B from an un-folded IIFE in the rebuilt output (cosmetic terser artifact, same semantics).

Per-bundler notes for reviewers:

• esbuild benefits most — it treats any module-scope call or member access as potentially side-effectful, so before this change a single type token dragged in the entire 6 KB token bundle.
• webpack partially compensated via its innerGraph analysis but still left 1.3–1.4 KB of unused tokens per import; now near-optimal.
• Vite (rolldown) sits between the two and needed both the sideEffects metadata and the purity fixes to reach ~optimal output (95–99 % reduction).
• Rollup was already near-optimal thanks to its deep statement-level purity analysis — it is unaffected by this change, which also serves as an independent regression check that the rewritten modules are semantically equivalent.
• @carbon/react component bundles (Button: ~50 KB min across bundlers) are unchanged — the package was already correctly configured; the wins land for anyone importing from the token packages directly or transitively (e.g. @carbon/elements).

Testing / Reviewing

• All @carbon/type and @carbon/layout unit tests pass (261 tests, 67 snapshots)
• Public export lists and token values deep-compared against the published packages — zero differences
• Bundle measurements above reproduce with single-import fixtures against the built es/ outputs

PR Checklist

• Reviewed every line of the diff
• Updated documentation and storybook examples
• Wrote passing tests that cover this change
• Addressed any impact on accessibility (a11y)
• Tested for cross-browser consistency
• Validated that this code is ready for review and status checks should pass

https://claude.ai/code/session_01TtzyHZUPcnLbC128bSn9Lr