Performance issues with excessive `noDup` and `isPerm` checks

[ymherklotz]

Currently compiling the loop example with mux and branch rewrites takes 1.2s, which is a bit too long for such a small graph. The culprits seem to be excessive noDup and isPerm checks, so either we need to prove that our maps never contain duplicates, or see if the checks can just be reduced.

The profiler report can be found here.