[Q21.5] Handle UTF-8 more robustly; pass through information about incorrect sequences

[Stephen J. Turnbull]

Aidan Kehoe writes:

 > You know that our internal string encoding is not exposed to Lisp,
 > except via CCL, right?  I don’t object to your asking to document
 > it, but I wonder what provokes the question.

I'm not talking about the internal encoding.  I want to know what
happens if you edit the buffer a buffer containing a representation of
non-UTF-8 stuff, and then use/save the result.  The AUCTeX processing
of TeX error messages described by David Kastrup would be a use case.
Another would be people trying to recover text from a core dump.

If the result is predictable and documented, that could be useful to
people who are deliberately working with buffers that do not contained
well-formed encoded test.