This test should now work in any database with UTF8 encoding, regardless
of the database's default locale.  The former restriction was really
"doesn't work if default locale is C", and that was because of not handling
mbstowcs/wcstombs correctly.