faster elaborator

[Matthew Fluet]

> pre code gen	196
> x86 code gen	492
> assembler	 32
> total		720
> gc		304

I've been improving the time and space for the x86 codegen phase.  I'll
send out an update probably on Friday; I need to change a few more things
on the allocate registers phase.  

> 	 allocateRegisters totals 318.090
> 	    toLiveness totals 136.530
> 	    toNoLiveness totals 0.0
> 	    Assembly.allocateRegisters totals 180.680
> 	       Instruction.allocateRegisters totals 97.390
> 		  pre totals 19.830
> 		  post totals 31.670
> 		  allocateOperand totals 29.970
> 		  allocateFltOperand totals 0.0
> 		  allocateFltStackOperands totals 0.0
> 	       Directive.allocateRegisters totals 20.620

I think I know what accounts for the discrepency between the
180.680 and the 97.390 + 20.620 = 118.01 and if I'm right, we'll recover
almost all of those 60+ seconds.  Also, some of the techniques I applied
to the simplify phase should cut down on the toLiveness time.