Instead of parsing every multibyte character twice, first with
parse_mbchar() and then with mbtowc(), just let mbtowc() do all
the work.  This makes searching for a fixed string twice as fast.

This also gets rid of four variables and lots of memory allocations.
(And, more importantly: it stops nano messing up the internal state
of the multibyte-to-wide character conversion, and thus would make
the calls to mbtowc_reset() superfluous.)