Summary

decimals_to_words only splits the first decimal point, so a dotted version like 0.7.0 came out as "0 point 7.0" — it expanded the first dot and left a stray full stop mid-number. This adds a versions pass so multi-part version strings read naturally.

Changes

• versions_to_words runs just before the decimal split: every . in a 3-or-more-part number becomes "point" (0.7.0 → "0 point 7 point 0", 1.20.3 → "1 point 20 point 3").
• Two-part numbers are left alone so ordinary decimals still read digit-by-digit (3.14 → "3 point 1 4"), and a trailing sentence stop is preserved (upgrade to 1.2.3.).

Testing

• pytest — adds unit tests for spelling out versions, leaving decimals and trailing stops intact, plus an end-to-end normalize_for_speech check; ruff check . clean.
• End-to-end: stackvox normalize "Upgraded to 0.7.0 today. Pi is 3.14." → Upgraded to 0 point 7 point 0 today. Pi is 3 point 1 4.

Related issues

None.