v1.2.1

v1.2.1 — Patch Release

Bug Fixes

• Version reporting: __version__ now reads from package metadata instead of returning hardcoded value
• Semantic checker cleanup: close() no longer crashes when model initialization was incomplete
• Correction output: generate_corrected_text no longer applies advisory-only (INFORM) errors to output text
• Grammar context matching: word rule mixin now correctly handles sentence_initial, after_verb, after_noun, and sentence_final_after_verb context values
• Grammar POS filtering: excluded_pos rules now apply correctly even when previous word has no POS tag
• Duplicate tense errors: removed legacy tense checker that duplicated TenseAgreementChecker output
• Negation checker: fixed lookahead producing duplicate errors on the same position
• Pipeline error handling: packaging failures now caught and rolled back properly
• Database VACUUM: runs outside implicit transaction to avoid SQLite errors
• Collocation mining: switched from fetchall() to cursor iteration to reduce memory usage on large databases

Other

• Correct README badges to reflect actual counts (4,658 tests, 74% coverage)
• Add compound_morphology.schema.json validation schema