Turns out we have to be a little more lenient when dealing with the
slightly hostile inputs in tests262-parser-tests.
This fix papers over some problematic situations by returning a fallback
range if we can't figure out exactly where an error occurred.
I've added a FIXME about returning nicer values.