BOMs in ascii or utf-8 documents is a bug, period.
Given that this bug is so prevalent, we decided to handle it,
but it is a kind of "implementation detail". Documenting it would sound
like a recognition that utf-8 BOMs could have some place in a reasonable