This updates our fnmatch port to more closely resemble the code
in CRuby. Specifically this work is intended to fix jruby#6193 but the
additional cleanup may fix other issues. Remaining work mostly
involves properly advancing the character pointers in the
presence of a multibyte encoding.

This fixes the remaining failurue in CRuby's test_fnmatch.

A bit of refactoring came along for the ride.

Part of work for jruby#6193

CRuby will blow by this code because even s == send will just
yield a null byte. We preemptively guard against p or s being
at their respective ends already.