Regression: The trailing Unicode sequences don't always show correctly for sequences...

Regression: The trailing Unicode sequences don't always show correctly for sequences ending normally.

The commit 85df83a6d846d575657016682c7014a09ac8af4e didn't handle all of the cases it needed to.
I forgot to check the normal behavior and ended up breaking normal behavior while fixing the exception cases.

The width_utf might be -1 for ASCII and performing the subtracting without handling -1 (and then casting it to unsigned) results in bad behavior.

Only handle incomplete character at the end of the stream when the character is actually incomplete.