Fix normalize_text handling of Nirugu and ZWJ

[satsragclaw]

Summary

• use one Mongolian word-run classifier for strict validation and normalize_text()
• keep Nirugu and ZWJ inside the same Mongolian run as their neighboring letters
• add regression tests for standalone and mixed-text normalization

Bug

normalize() accepted Nirugu and ZWJ as word characters, but normalize_text() split runs at those characters. Splitting at a joiner changes neighboring letters' isol/init/medi/fina positions, so the same word could receive a different canonical result depending on which API was used.

Verification

• python3 -m compileall -q mongol_norm tests
• python3 -m unittest discover -s tests -p 'test_*.py'
  • 145 tests passed
  • 177/177 core-hud direct cases
  • 3507/3507 eac-hud direct cases, with 5 documented UTN xfails skipped
  • 225/225 CORE-HUD round trips
  • 3513/3513 EAC-HUD round trips
  • 2237/2237 prefix-stability pairs
  • 1993/1993 shape groups converge