arrow

ARROW-11944: [Developer] Fix archery's comparison of cached benchmark runs

Also added a unit test and some notes to `benchmarks.rst`.

My own preferred benchmarking workflow is to exclusively output run results to json and run `archery benchmark diff` against those. I'm not sure that automatic (re)generation and rebuild of the cmake build directory is a good default; IMO a new user will most likely be trying to iterate rather than make a single shot run and should be protected from staring at a release build log after every change. Is there enough guidance in `benchmarks.rst`?

@dianaclarke @edponce

Closes #9712 from bkietz/11944-Achery-benchmark-diff-reg

Authored-by: Benjamin Kietzman <bengilgit@gmail.com>
Signed-off-by: Benjamin Kietzman <bengilgit@gmail.com>