Fixed a bug in PyUnicode_Count() which would have caused a
core dump in case of substring coercion failure.

Synchronized .count() with the string method of the same name
to return len(s)+1 for s.count('').