x86 update

[Matthew Fluet]

> We're back to a repeatable bug in the codegen with -native-live-transfer true.

I guess that's progress.

> In the next message, I attach a (working :-) snapshot that demonstrates the bug.
> Here is a log of the failed compile.

What was the compile sequence?  In particular, what were the options used
to compile the G1 mlton that is doing the compile in the log?  If the G1
mlton is a -native-live-transfer false, then it should just be a bug in
the computeLiveTransfer phase (or maybe something using that information);
if the G1 mlton is a -native-live-transfer true, then it could be either a
buggy computeLiveTransfer phase or I could be producing buggy assembly.
I'd suspect the former, only because it would seem to be an amazing
coincidence that the bug in the assembly happened to fall in the
code for the x86-backend.