Fix stale glossary-sensitive output adoption

[cursor]

Bug and impact

Selective re-translation could silently adopt existing output_chunk*.md files that had no run_state.json record, even when the current source chunk selected glossary terms. If the output was produced before a glossary edit, record-all stamped the stale translation as current and future plans reported it unchanged, causing persistent terminology corruption in final books.

Root cause

run_state.plan() treated every untracked existing output as record-only by default, without checking whether the current glossary would be injected into that chunk. glossary.term_hash() also ignored aliases, so alias-only prompt changes for already-selected terms did not invalidate prior records.

Fix

• Re-translate untracked outputs whose source chunks select glossary terms; continue record-only adoption for glossary-free chunks.
• Include sorted aliases in per-term hashes when aliases are present.
• Add regression coverage for untracked glossary-sensitive outputs, record-all adoption, and alias hash changes.
• Update SKILL/README docs to describe the safer adoption behavior.

Validation

• python3 -m unittest tests.test_run_state tests.test_glossary -v
• python3 -m compileall scripts tests
• python3 -m unittest discover -s tests -p 'test_*.py' -v (209 tests, 3 skipped)
• CLI repro now keeps stale glossary-sensitive chunk0001 in translation_chunk_ids before and after record-all.
• Standard Alice mock pipeline generated HTML/DOCX/PDF successfully; EPUB failed only due the documented Ubuntu Calibre 7.6.0 packaging bug.