SSA backend

[Matthew Fluet]

Well, that was deceptively simple.  There was one error in the porting of
the translation of overflow primitives from x86-mlton.fun to
x86-translate.fun.  (The commuting operations might switch src1 and src2,
but the auxilary function check had src1 as a free variable; hence, it was
always bound to the original src1, not necessarily the switched.)
Serendipitously (sp?), it was the failure to add the comment to the
output assembly for -native-commented n (n >= 1) that made me look
carefully at the overflows.

Anyways, I got through all the regressions and am looking at a
self-compile now.  If the resulting G1 can get through the regressions and
benchmarks, I'll check it all in.