string_decoder: align UTF-8 handling with V8

V8 5.5 changed how invalid characters are handled and it now appears
to follow the WHATWG Encoding standard, where all of an invalid
character's bytes are replaced by a single replacement character
(\ufffd) instead of replacing each invalid byte with separate
replacement characters.

Example: the byte sequence 0xF0,0xB8,0x41 is decoded as '\ufffdA' in
V8 5.5, but is decoded as '\ufffd\ufffdA' in previous versions of V8.

PR-URL: https://github.com/nodejs/node/pull/9618


Reviewed-By: Ali Ijaz Sheikh <ofrobots@google.com>
Reviewed-By: Ben Noordhuis <info@bnoordhuis.nl>