citrus: Avoid invalid code points.
The UTF-8 decoder should not accept byte sequences which decode to
unicode code positions U+D800 to U+DFFF (UTF-16 surrogates).[1]
Contrary to the original OpenBSD patch, we do pass U+FFFE and U+FFFF,
both values are valid "non-characters" [2] and must be mapped through
UTFs.
[1] http://www.cl.cam.ac.uk/~mgk25/unicode.html#utf-8
[2] http://www.unicode.org/faq/private_use.html
Reported by: Stefan Sperling [1]
Thanks to: jilles [2]
Obtained from: OpenBSD
9fe5eca952