RSSA

[Matthew Fluet]

> regressions and benchmarks.  Matthew, if you could try pushing the
> changes into the x86-codegen, that would be great.

I started looking at some of the changes.  One question: are you going to
make array inialization explicit in RSSA (and/or Machine)?  If you don't,
then we need the live variables at an array allocation to fill in the live
variables that are live around the initialization loop.  If you do, bear
in mind that that will be a new loop where a signal limit check will be
inserted.