Preserve multibyte UTF-8 characters when obfuscating.
This version uses the unicode 'replace' option, so non-UTF-8
encodings will still be mangled. Luckily, there aren't any
of those.
# should print nothing# should print ? b/x? b/x# should print A b/xA b/x# should forget b/xforgetting b/x# should print nothing