Preserve DOCX run boundaries on PM round trip

[aldrinjenson]

Summary

• Preserve source DOCX run boundaries through no-op toProseDoc -> fromProseDoc round trips.
• Restore empty source runs and split ProseMirror-merged same-mark text only when the paragraph text and mark stream still match the imported source.
• Add regression coverage, a patch changeset, and regenerated API snapshots for the new internal paragraph attr.

Testing

• bun run format
• bun test packages/core/src/prosemirror/conversion
• bun run typecheck
• bun run check:parity
• bun run build:packages
• bun run api:check

Refs #830

---

^{Need help on this PR? Tag /codesmith with what you need. Autofix is disabled.}

[vercel]

@aldrinjenson is attempting to deploy a commit to the EigenPal Team on Vercel.

A member of the Team first needs to authorize it.

[eigenpal-release-pal]

All contributors have signed the CLA ✍️ ✅

_{Posted by the CLA bot.}

[aldrinjenson]

Update after corpus verification: the original version only restored boundaries after toProseDoc, but the real-doc gap was earlier. This branch now also stops parseParagraph from applying consolidateParagraphContent during DOCX parse, so adjacent same-rPr <w:r> elements and leading empty runs survive into the document model before the PM round trip.

Sanity check on the #830 attachment: raw document.xml has 139 body paragraphs, parsed model has 139, and the paragraphs with 7 raw <w:r> runs now parse as 7 model runs.

Additional validation after the parser fix:

• bun test packages/core/src/docx packages/core/src/prosemirror/conversion
• bun run typecheck
• bun run check:parity
• bun run api:check

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docx-editor	Ready	Preview, Comment	Jun 16, 2026 3:02pm

[jedrazb]

thanks for taking this on, the approach is solid. i like that it fails closed, and i couldn't get it to corrupt or duplicate content. requesting changes for a few things before merge:

1. needs a rebase (behind main). since the parser de-consolidation is pipeline-wide, not just round-trip, please run the full playwright suite after rebasing, not only the conversion unit tests.

the rest are inline. nice work on the changeset, the regenerated api snapshots, and the stale-after-edit coverage.

[aldrinjenson]

Pushed e97ddd5 addressing the review comments.

Validation run:

• bun test packages/core/src/prosemirror/conversion/__tests__/run-boundary-roundtrip.test.ts passed
• bun test packages/core/src/docx packages/core/src/prosemirror/conversion passed
• bun run typecheck passed
• bun run check:parity passed
• bun run api:extract && bun run api:check passed
• pre-commit repeated typecheck/parity/API extraction successfully
• bun run test:e2e was attempted as requested, but the full Playwright suite is not green in this checkout: 793 passed, 5 skipped, 213 failed. The failures are broad UI/e2e issues (toolbar/dropdown selector timeouts, visual snapshot differences, table toolbar actions, find/replace, scroll-to-page, etc.) rather than localized failures from this run-boundary change.