Analyze and improve pre-validation steps

[matt-sheets]

The pre-validation steps introduced here: #128
are not very optimized and can be improved.

Currently we have what could potentially be a large clone which sizable policy (which ref policy 3 will become) in a while loop that may take several iterations to resolve. I initially did this to avoid mutability & ownership issues. This can be seen in search_for_recursion

I assume there are also general improvements that can go into this function as well.

[dburgener]

Can you add a little more specificity? You're talking about performance, and specifically the repeated large clone() I imagine?

[dburgener]

I did a first pass at some optimizations here. Converting to a BTreeSet and avoiding removals inside a Vec was a huge (50%) win. Removing the clone() showed no performance improvement. My guess is that maintaining the list of things to remove cost enough that it canceled out the clone() removal. I'm going to submit a PR with the BTreeSet conversion and that speedup, but I'll leave this open for now.