Better automate comparing two Miri revisions with ./miri bench

[RalfJung]

Currently to figure out the impact of a Miri change on our benchmarks, one has to run ./miri bench twice and then manually compare the results. We should have a way to automate that... but I have no clue how to do that.

hyperfine supports the case of having multiple benchmark commands and then it will print comparisons between them, but that doesn't work for us since we want to gather the two measurements separately, and then compare after the fact. We probably want ./miri bench to dump the results into a file, and then have a way to compare two such files. But does hyperfine even have a way to dump the results in machine-readable form? And can we avoid having to implement the comparison code ourselves?