Fix stale glossary-sensitive output adoption

[cursor]

Bug and impact

Untracked output_chunk*.md files were record-only adopted by default even when their source chunks selected current glossary terms. In a resumed or manually edited temp dir, stale translations could be stamped into run_state.json as current, causing later runs to skip required re-translation and silently preserve wrong terminology in final book outputs.

Root cause

run_state.plan() handled every existing output without a state record as record unless --retranslate-untracked was set. Separately, glossary.term_hash() ignored aliases, so alias-only glossary edits for already-selected terms did not invalidate recorded chunks.

Fix

• Re-translate untracked outputs whose source chunks select current glossary terms; continue record-only adoption for untracked chunks with no selected terms.
• Include sorted aliases in per-term hashes while preserving no-alias hash compatibility.
• Update SKILL/README docs and add regression coverage.

Validation

• Reproduced the pre-fix CLI scenario where stale output_chunk0001.md was recorded as unchanged.
• python3 -m unittest tests.test_run_state tests.test_glossary -v passed (68 tests).
• python3 -m unittest discover -s tests -p 'test_*.py' -v passed (209 tests, 3 skipped).
• python3 -m compileall scripts tests passed.
• Post-fix CLI repro keeps glossary-sensitive chunk0001 in translation_chunk_ids through record-all, while record-only adopting glossary-free chunk0002.
• Baseline standard-alice mock pipeline generated HTML/DOCX/PDF; EPUB failed only for the documented Ubuntu Calibre 7.6.0 issue.