self compile problems with SSA simplifier

[Stephen Weeks]

> All of the pre-SSA passes are of the usual speed, but the SSA
> optimizations are sometimes more than 100X too slow.  See,
> e.g. removeUnused1.  The problem isn't too much live data, since GC%
> is roughly normal.  Also, since it happens to all SSA passes, I'm
> inclined to blame the shrinker, or some kind of property list
> problem.  Anyways, I'm looking into it, but feel free to do so as
> well.

I found some debugging code in the shrinker that got left on.  I've
turned it off and am re-running the tests.

I forgot to mention the other problem, which was that commonSubexp
raised Overflow when self-compiling.  I'm not sure whether it's that
pass or the shrinker.