fix(i18n): skip regex fragments in the translation check

[Caldalis]

Problem

npm --prefix frontend-enterprise run i18n:check currently exits 1 on main:

Missing English translations (1):
- src/pages/chat/chatHelpers.tsx:1707:string: "(?:参考来源|参考资料|引用来源|资料来源)"

The literal was introduced by 120ce01 (fix(chat): deduplicate knowledge citations). It is the citation-heading pattern that
stripTrailingCitationSummary compiles into a RegExp — not UI copy. check-i18n.cjs flags every Chinese string literal that has no en.json entry and cannot tell the two apart, so it reports the regex as a missing translation.

Change

Add the fragment to the ignoredFragments allowlist that already exists in check-i18n.cjs, plus a short comment recording what that list is for.

Why not add it to en.json

localizeTextNode / localizeAttribute in src/i18n/index.tsx perform DOM-level replacement over text nodes and attributes; they never touch JS values. This fragment cannot reach the DOM, so an en.json entry would be a permanently unmatched row and would mix non-UI strings into the translation catalog. ignoredFragments exists for exactly this case and already carries after:content-['展开'] (a CSS content value).

Risk

Low. Only the validation script changes — no runtime code, no build output, and en.json is untouched.

Tests

Run on this branch:

• npm --prefix frontend-enterprise run i18n:check — passes (3235 translations)
• npm --prefix frontend-enterprise test — 50 files / 243 tests passed
• npm --prefix frontend-enterprise run build — passes (includes tsc -b)
• npm --prefix frontend-enterprise run config:check — passes

UI validation

Not applicable: this changes only a script under frontend-enterprise/scripts/ and affects no route or rendered output, so there are no routes/roles to verify and no screenshots.