This brings the two utf-8 decoders (bulk + incremental) in line.
Technically, either behaviour was correct, since the utf-8 spec
demands incomplete utf-8 be handled, but does not specify how.
Unicode recommends that "the maximal subpart at that offset
should be replaced by a single U+FFFD," and with this change we
consistently do that. More details + spec references in the bug.

BUG=chromium:662822

Review-Url: https://codereview.chromium.org/2493143003
Cr-Commit-Position: refs/heads/master@{#41025}