toke.c: Don't force \N{} into utf8 unnecessarily
regcomp.c no longer requires everything containing \N{} to be in UTF-8.
I'm not sure of the exact commit; it might even have been in 5.16. But
it was done by the time
86e88272fdabc40e3b168a3cc43af5e86284d01b was
done.
Therefore we can remove the temporary code that forced utf8, and replace
it with code that handles the non-utf8 case.
Note that outside patterns, \N{} still forces utf8. This is so that
Unicode semantics will be imposed on the string it resides in, no matter
how it is used. Patterns have a flag that indicates Unicode semantics,
so don't need to be in utf8.