limit check insertion via loop forests

[Matthew Fluet]

> > B.t.w., there is one complication with Steve's represenation of loop
> > forests, which is that all of the nodes that aren't in any loops are
> > in the same notInLoop vector (at the top level) are effectively in the
> > same loop.  Not a big deal, when setting up the equivalence classes, we
> > just need to specially treat those in the top notInLoop vector.
> 
> Actually, this was a big deal with the original definition of L, because
> every label that followed a non-loop label would be put into L.  But, as I
> said, it's just special treatment of those in the top level.

Sorry, no.  Steve's right, it does make sense to put all non-loop labels
in the same "loop" equivalence class.  Interestingly, though, the set of
"extra" nodes will be a super set of Headers (again, I'm thinking
Steensgaard loop headers), because loop headers always occur at the
boundary between one loop class and another.  Something to keep in mind if
anyone comes up with an intuitive way of getting the double loop example I
proposed with a minimal set of limit checks.