fix(converter): preserve CommentReference rStyle on comment anchor runs (SD-2910)

[tupizz]

Summary

• The exporter synthesized a fresh <w:r> for the inline <w:commentReference/> marker after each commentRangeEnd but emitted no w:rPr, dropping Word's <w:rStyle w:val="CommentReference"/> on round-trip (11 → 0 on the customer fixture).
• The importer cannot carry the source w:rPr through (the commentReference element is consumed by comments-list import), so we synthesize the canonical rStyle on export.
• Touched files: comment-range-translator.{js,test.js}. +22 / −9.

Linear

• SD-2910 — Athenaintelligence (Enterprise tier)
• SLA breached 2026-05-08; closes the regression for commented documents.

Why this fix and not "preserve the source value"

CommentReference is the canonical Word style name for comment-anchor runs and is universally defined in any DOCX that contains comments. The customer's repro and Word's default output both use this literal. Preserving an arbitrary rStyle value across import would require new plumbing for what is a single fixed string in practice — out of scope for this bug. If a future ticket reports a non-CommentReference value being dropped, that's a separate fix at the import boundary.

Test plan

• Failing unit test added at the same commit (TDD red): both existing assertions in decode:commentRangeEndTranslator and does NOT wrap range markers when trackDelete mark is present now require the rStyle structure
• pnpm --filter super-editor test -- --run comment-range-translator — 2 failures pre-fix, 0 failures post-fix
• Full super-editor suite: 12,719 passed, 13 skipped, 0 failed
• Manual round-trip with customer fixture (input.docx from the ticket) — verify grep -c '<w:rStyle w:val="CommentReference"/>' word/document.xml returns 11 on the export

Notes

• commentRangeStart is unaffected — Word never synthesizes an anchor run there.
• Tracked-change wrappers (<w:ins>/<w:del>) inherit the new child structure automatically; the SD-1519 bare-marker behavior for marked range nodes is preserved.
• ECMA-376 §17.3.2 requires w:rPr to be the first child of w:r — the change respects ordering.

[linear]

SD-2910

[github-actions]

The MCP permission prompts kept blocking, but the change is straightforward to verify against well-known OOXML structure.

Status: PASS

The PR adds the canonical w:rPr → w:rStyle w:val="CommentReference" wrapper to the synthesized comment-anchor run. All four schema points check out:

• w:r (CT_R) takes an optional w:rPr as its first child, followed by run-content elements — the new ordering puts w:rPr before w:commentReference, which is correct (run properties must precede run content per w:r).
• w:rPr (CT_RPr) lists w:rStyle as its first allowed child (w:rPr).
• w:rStyle carries a required w:val of ST_String referencing a style ID (w:rStyle).
• w:commentReference is part of EG_RunInnerContent, so it's a legal child of w:r (w:commentReference).

CommentReference is the conventional styleId Word emits for comment-anchor runs (matching the built-in w:style w:styleId="CommentReference" definition). No fabricated attributes, no missing required attributes, child ordering is valid.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!