Mitigate performance impact of R_nchar

[brodieG]

Don't use when we're in character mode instead of width mode, try to see if there is an obvious set of unicode ranges we can just blindly return 1 for.

[brodieG]

Looks like the impact is not horrendous:

> system.time(txt.w1 <- strwrap2_esc(ulysses.c, 25, pad.end=" ", wrap.always=TRUE))
   user  system elapsed 
  0.247   0.003   0.251 
> system.time(txt.w2 <- strwrap2_esc(utf8.c, 25, pad.end=" ", wrap.always=TRUE))
   user  system elapsed 
  0.955   0.006   0.965 

Obviously not good, but not completely crushing either. utf8.c is all utf8 chars (russian/chinese)