a more efficient loopForestSteensgaard

[Stephen Weeks]

I just checked in code that computes loopForestSteensgaard more
efficiently.  It also returns a loop forest represented in the
following way.

      datatype t = T of {loops: {headers: Node.t vector,
				 child: t} vector,
			 notInLoop: Node.t vector}

I'd like to consider replacing the return type of
loopForestSteensgaard with the above, but I don't understand the
relationship well enough with what's currently there or with how it's
used in x86LoopInfo.  Matthew, could you take a look at the code, make
sure you agree it is computing the same loop forest (and is more
efficient), and give me your thoughts on converting to the above as
the representation of the loop forest?  Thanks.