* This pass was written way back when ... when we were concerned
  about interp perf on its own, and this was running before
  local opt pass and could disable some local opts because of
  introducing RAW hazards. So, I had added some temp-specific local
  opts in there which were fragile (and over time, code and comments
  had diverged).

* Since we swapped the ordering to run local opt pass before this one
  (before disabling it), we no longer need to run the fragile local
  opts all over again in this pass (which also happened to have been
  buggy!).

* With these fixes, we can once again re-enable this pass, if we
  want to without breaking tests. Let us make that decision separately
  if this pass is even worth it, and let us measure how much perf.
  this even gains us. Does it produce any memory savings? If so,
  how significant. If this is not useful perf-wise, we should just
  kill this code since we are more interested in JIT perf now.