Open source benchmark runner script

We have nice benchmark runner script internally that's unfortunately tied into a few small internal infrastructure things (e.g. Blaze, how to get a clean version of the tree ...) -- we should factor out the internal detail bits (e.g. a "blazel == blaze + bazel", clean checkout flag) so we can run this in OSS as well. We can do some simple things as replacements, pass a flag to point at the clean git checkout to use as a baseline.