Our IcuCaseFold() fallback using strxfrm() is not actually case
insensitive.  This commit fixes the problem by switching to
strcasecmp().  That function is not guaranteed to support UTF-8, but
it's the best we can do in this sparse situation.

Closes #111