refactor(core): extract shared sanitizeUnicodeText from session-name/…

[chinawch007]

Summary

• foreign-session.ts:sanitizeForeignText duplicated the session-name.ts unicode pipeline and the two had drifted — session-name was missing 8 code points that foreign-session (and its own FOREIGN_UNSAFE_CHARS id guard) already covered: U+061C, U+200E, U+200F (bidi marks) and U+2060–U+2064 (invisible format chars).

• I extracted the pipeline into a shared pure helper sanitizeUnicodeText(text, {maxCodePoints, truncatedSuffix}) in a new packages/core/src/text-sanitize.ts leaf module, which now holds the three character-class regexes as the single source of truth. The character classes take the union — the stricter foreign-session coverage — so session-name picks up the 8 previously-missing code points as a pure coverage gain (they never appear in legitimate CJK/emoji text, so no false positives).

• Because the two callers disagree on truncation behavior and both are pinned by existing tests, I parameterized the suffix: foreign-session appends … (the default), while session-name passes '' to preserve its silent-cap behavior. The helper boundary is clean-and-truncate only; the type guard, empty-string policy, and Result-vs-string return shape stay with each caller, so both export signatures are unchanged and no downstream caller needs edits.

• foreign-session.ts:sanitizeForeignText is now a thin wrapper around the helper, and session-name.ts:normalizeUserSessionName calls it with truncatedSuffix: '' (net −59 lines). I also exported sanitizeUnicodeText from index.ts and added one session-name test case locking in the newly-covered bidi marks and invisible operators so they can't silently drift back out of sync. FOREIGN_UNSAFE_CHARS and isSafeForeignId (the verbatim-rendered id guard) are a separate concern and left untouched.

Refs: #1404

Verification

• npm run format:check
• npm run lint
• npm run typecheck (packages/core) — passed, 0 errors
• npm --workspace @maka/core run build — passed
• npm --workspace @maka/core run test:dist -- 782 passed

[Astro-Han]

Thanks for this — I verified the extraction at character level against the pre-PR pipelines: the foreign call site is byte-identical (same regex sets, same ordering, same '…' suffix, same non-string handling) across a 60+ input battery, and the session-name site differs on exactly the 8 documented code points (ALM/LRM/RLM → space, U+2060–U+2064 → removed) — matching the commit message precisely, with the new test cases failing on the old code as they should. The scope is right too: UNSAFE_REVIEW_CHARACTER and the other sanitize-ish code are deliberately different guards, not candidates for unification. All 33 tests pass on the PR sources.

Three optional notes:

• The "one source of truth" claim for FOREIGN_UNSAFE_CHARS now spans two files (the sanitizer's classes live in text-sanitize.ts); nothing enforces that the id-guard set stays equal to the sanitizer union. A small contract test (every code point the sanitizer accepts is rejected by the guard) would pin it.
• U+FEFF and U+200C removal through the shared function are untested (pre-existing gap, worth two lines in the new union-coverage test).
• The session-name change is user-visible (marks render as spaces) — it's pinned by the new test, just worth an explicit mention in the PR body/release notes.

Nothing blocking — happy to approve.