Fuzz for nondeterministic behavior

We should fuzz for non-deterministic behavior in XLS. It shouldn't be hard to tweak the fuzzer to do this. The fuzzer would generate a sample like it normally does, but would then run the sample (generate IR, optimize, schedule, etc) more than once. The generated artifacts (IR , optimized IR, verilog, etc) between the different runs would then be compared against each other.