[MLton] performance improvements for redundant arguments optimization

[Stephen Weeks]

Luke, I checked in some improvements to redundant.fun.  It is now
linear, not quadratic, in space usage.  Hopefully, hamlet and
model-elimination can now compile with full flattening.  I'd also be
interested in seeing a self compile.  Also, since the full flattening
is tickling slowdowns in some other simplifier passes (at least for
hamlet and model-elimination), I'd like to hear what those are.

BTW, I ran all the benchmarks {,-drop-pass redundant} and observed
that the redundant pass has no effect.  Maybe it will with more
flattening, though.