Errors, yet again

[RWK at MIT-MC (Robert W. Kerns)]

The system should not limit the classes of errors possible to the ones that
the system generates (I.e. Wrong-type-argument, wrong-number-args,
unbound-variable, undefined-function, etc.).  There should be a way for the
user to signal and provide handlers for errors in exactly the same manner as
the system does.

For example right now in MACLISP, the ERROR function's third argument
could be a symbol which is considered to be the error-class, and it's value
to be the handler, and the value returned to be the value returned from the
handler, etc.  I think that this would be a useful and compatible change
(if the arg was a fixnum it would be as before).  I don't think this is the
final word on how to win though....