Evidently normalizing any whole string is too slow for completion,
so we need to even out case and composition while iterating through
the strings.
The decompositing version of katze_utf8_stristr is disabled, since
it is too slow, and an ascii only version is used, for now.
A unit test 'compare' is added that solely measures performance of
katze_collfold and katze_utf8_stristr.
The test was modified slightly because we are not filtering
out trailing slahses anymore. We are not quite there still
when it comes to treating URIs as we should but for all
practical purposes it should be quite a bit faster.
The most annoying shortcoming still is that umlauts or any
special characters won't match at all.
