bug#12758: letf no longer allows unbound variables

[Stefan Monnier <monnier@iro.umontreal.ca>]

> Evaluating this returns `99', and leaves `b' unbound. This behaviour is
> documented in cl.texi:

>    In most cases, the @var{place} must have a well-defined value on
>    entry to the @code{letf} form.  The only exceptions are plain
>    variables and calls to @code{symbol-value} and @code{symbol-function}.
>    If the symbol is not bound on entry, it is simply made unbound by
>    @code{makunbound} or @code{fmakunbound} on exit.

I think this was a bad idea, so it indeed doesn't work that way any
longer; more specifically:

- W.r.t symbol-function, this is still true for `letf' but not for
  `cl-letf'.
- W.r.t symbol-value, this is not true any more neither of `letf' nor or
  `cl-letf' (I could change that for letf but in the absence of
  a bug-report pointing to pre-existing code that depends on this
  behavior I'd rather not).
- For (letf ((b 4)) ...) this is still true, because it expands to (let
  ((b 4)) ...).  But for (letf ((b)) ...) it isn't because that expands
  to (let ((b b)) ...).
  When `b' is a lexically-scoped variable, we really can't "fix" it
  because lexical variables don't have a notion of "unbound".


        Stefan