#8271: the utf-8 decoder now outputs the correct number of U+FFFD  characters when used with the "replace" error handler on invalid utf-8 sequences.  Patch by Serhiy Storchaka, tests by Ezio Melotti.