Revamp ibcmp_utf8 for performance and clarity

Revamp ibcmp_utf8 for performance and clarity

I had a hard time understanding how this routine worked; there were no
comments.  In figuring it out, I discovered it could be made more
efficient.  This routine is called over and over in the innermost loops
in regex matching, so efficiency is a concern.

Setup is done once before the main while loop so that it now has two
conditions instead of eight.  The loop was rearranged slightly to be
smaller and a couple of unneeded assignments to temporaries were
removed, and recomputation of some values was avoided.  Several other
small efficiency changes were made.

Several asserts had been commented out, saying that they make tests
fail.  But they no longer do, at least on my platform.  There was a
reason that they were asserts to begin with, and that is they denoted an
insane or trivial condition.  Apparently there have been fixes to the
other code calling this, so I re-enabled them.

The names of several variables were changed to be less confusing; hence
f1 means the fold buffer for string 1 whereas it used to mean its goal,
which is now g1.

The leading indent was changed from 5 to 4 blanks.  I made enough
other changes that I didn't submit this as a separate commit