(encoded-kbd-self-insert-utf-8): Catch and recover from case when the bytes we thought we were reading turn out to be something else entirely, such as latin-1 chars from quail. See bug#396.