Reported by Fredrik Lundh on python-dev.

The conversimple() code that handles Unicode arguments and converts
them to the default encoding now calls converterr() with the original
Unicode argument instead of the NULL returned by the failed encoding
attempt.