To avoid byte/character mixture, getarg('c') must only accept a byte 
string of 1 byte and getarg('C') only an unicode string of 1 
character.

Impacted methods:
- datetime.datetime.isoformat(sep), array.array(type, data): don't 
accept byte anymore
- msvcrt.putch(char), msvcrt.ungetch(char), <mmap 
object>.write_byte(char): don't accept unicode anymore

I tried runtests.sh: only the 3 bytes.{center,ljust,rjust} tests have 
to be changed, all other tests are ok.

Related issues: #3446, #5391.

unicode methods center(), ljust(), rjust() use a callback (convert_uc) 
to get an unicode character (unicode string of 1 character). If 'C' is 
fixed, convert_uc() callback can be removed.

Note: #5391 and #5410 are depends on this issue.

good catch.  the patch looks correct to me.

Fixed in r71013.

Ooops, I read converterr("a unicode character", ...), but it should be 
converterr("an unicode character", ...) ;-)

Thanks benjamin.peterson, it will help other issues like #5391 and 
#5410.