Problem: like_range() returned wrong ranges for contractions (like 'ch' in Czech').
Fix: adding a special code to handle tricky cases:
- contraction head followed by a wild character
- full contraction
- contraction part followed by another contraction part,
  but they are not a contraction together.