Fix stale adoption of glossary-sensitive untracked outputs

[cursor]

Bug and impact

Untracked existing output_chunk*.md files were record-only by default, even when the current glossary selected terms for that chunk. After a glossary edit or legacy adoption, stale translations could be recorded under the new glossary state and then skipped forever, silently shipping incorrect terminology.

Root cause

scripts/run_state.py plan() classified missing run_state.json records as record before running glossary selection/hash checks. glossary.term_hash() also omitted aliases even though aliases are part of the prompt table.

Fix

• Re-translate untracked outputs that would receive current glossary terms; keep glossary-free untracked outputs record-only for resumability.
• Include sorted aliases in term hashes when aliases are present.
• Update SKILL and README docs for the safer default behavior.

Validation

• Pre-fix repro showed chunk0001 with Tai incorrectly planned as record_only; post-fix repro plans it for translation and keeps only glossary-free chunk0002 record-only.
• python3 -m unittest tests.test_run_state tests.test_glossary -v (69 tests)
• python3 -m unittest discover -s tests -p 'test_*.py' -v (210 tests, 3 skipped)
• python3 -m compileall scripts tests
• Standard Alice mock full-pipeline generated HTML/DOCX/PDF; EPUB failed only due the documented Ubuntu Calibre 7.6.0 packaging bug.