Issue: LOAD-TIME-EVAL

[Kent M Pitman <KMP@STONY-BROOK.SCRC.Symbolics.COM>]

My notes mention that Haflich brought up the issue of dead code removal.
That is, does
  (LAMBDA () (IF T 'ALWAYS (LOAD-TIME-VALUE (F))))
-ever- have to evaluate the given expression (given that the value
might never be used).

The issue was passed 12-2 in spite of this objection, but I think we
might still want to clarify this issue if it's not clear already [I
didn't have time to check yet.]

I think it would be most consistent to say this might not ever get 
evaluated so that dead-code removal was legit here.  (In general, I
think it's permissible for LOAD-TIME-VALUE forms to do a side-effect,
but I don't think the -purpose- of doing a LOAD-TIME-VALUE is to get
a side-effect.  The side-effects are usually incidental to getting
the value.)