build times for each successful merge by bors. The Windows
tests are much faster because it runs a stripped
testsuite. The information coming from the buildbots is
very noisy, so most spikes are purely an artifact of that, not
a performance regression (or improvement) due to the
associated pull request.