Support profiling comparisons

[cstjean]

TODO:

• Write benchmark_median and benchmark_minimum convenience functions. Update: this became median circ benchmark

The main issue left is that the loaded trace has a bunch of ReconstructedType, and we need to handle those. Either write a custom compare_benchmark(minimum, tr) that reruns the benchmarks, or have map use @ignore_errors. I'm leaning towards the latter. We might then write map(ratio, tr1, tr2). This will require writing ratio(tr1, tr2) = ratio(value(tr1), value(tr2)).

[cstjean]

This is mostly done, but I want to do something better about errors when reloading a saved benchmark. I'd like to remove them and warn, which is the best I can do.