infinite loop on array.sml

[Matthew Fluet]

> > Yeah, I get that.  It's somewhere in main_0, so it's a pain to figure out.
> > I mean to ask, could the new dominator version that doesn't care about
> > the graph being connected and single rooted result in a dominator tree
> > that has a cycle?
> 
> Maybe there is a bug in dominator computation causing the loop, but if
> it terminates, we know from the tree type
> 
>       datatype 'a t = T of 'a * 'a t vector
> 
> that the tree really is acyclic.

O.k.  In any event, I'm fairly certain that after commonSubexp, there will
be some dangling blocks in the cfg; particularly down the failure branches
of repeated arithmetic computations.