fix(collaboration): render cross-references for late-joining clients (SD-2975)

[luccas-harbour]

Summary

Fixes SD-2975: cross-references (and citations) not rendering for collaborators who join an active room. Word-imported crossReference/citation field nodes carry cached result runs in their inline content, but the ProseMirror schema declares them as leaf atoms — so when y-prosemirror hydrated the shared Yjs XML for a late joiner, the cached children violated the schema and the node failed to materialize.

• Strip cached children from atom nodes during Yjs hydration. New normalize-yjs-fragment.js walks the supereditor XML fragment and clears children of crossReference/citation XmlElements. Runs before yXmlFragmentToProseMirrorRootNode and via observeDeep for incoming remote events. Runs inside a Yjs transaction with a private origin symbol so the normalizer ignores its own events.
• Fall back to resolvedText on export. Once cached content is stripped, the editor still needs to round-trip something to DOCX. New buildFieldResultRuns shared helper rebuilds a w:r with the node's resolvedText attr when node.content is empty, applying xml:space="preserve" for boundary whitespace. Both crossReference-translator.js and citation-translator.js now route through it.
• Editor lifecycle: OpenOptions accepts a fragment (Y.js XmlFragment) so callers can hydrate directly from a seeded room; the option is cleared on close to prevent stale fragments leaking across reopens. Pre-hydration normalization is also called in Editor.ts.
• Tests cover hydration of a seeded crossReference from a Yjs room, normalizer cleanup on extension teardown, scoping of the normalizer's own transactions, citation atom-content normalization, and export fallback when cached content has been stripped.

[linear]

SD-2975

[github-actions]

I wasn't granted permission to call the ecma-spec MCP tools, so I'm reviewing against ECMA-376 from working knowledge. The code changes are small enough to assess directly.

Status: PASS

The new buildFieldResultRuns helper in build-field-result-runs.js emits a standard w:r containing w:rPr followed by w:t, which is exactly the shape ECMA-376 prescribes for CT_R (w:r, w:t). Specifically:

• Child order is correct — w:rPr precedes the run content (w:t), as required by CT_R.
• xml:space="preserve" is conditionally applied to w:t only when resolvedText has leading/trailing whitespace, matching the W3C convention OOXML inherits.
• w:rPr is always emitted (even with empty outputMarks), but an empty <w:rPr/> is valid per the schema.
• \u200e (LRM) in the test fixture is correctly not treated as whitespace — it's a directional formatting char, so omitting xml:space="preserve" is right.

The two translator changes (citation-translator.js, crossReference-translator.js) just delegate to the shared helper; their existing call to buildInstructionElements for the REF/citation instructions is unchanged, and the w:fldChar/w:instrText framing they emit elsewhere is not touched here. No new attributes or elements are introduced — the PR only adds a fallback rendering path using already-valid OOXML primitives.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2a51835562

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!