This is part of CPython performance work that @gvanrossum and I are starting. We want to use pyperformance to run our benchmarks (and profiling, but that's a separate issue). As there isn't a mechanism to plug in custom benchmarks, I'm putting up this PR.

I suppose there's the question of whether this is an appropriate benchmark for the suite. I think so. It represents a workload, based on concrete usage patterns for a widely used CLI app, that does not seem to be covered by other benchmarks. Note that at the moment I'm cheating a little by running tests out of the azure-cli test suite. However, I expect the tests exercise the code in a very similar way to common usage of the actual application. If there are concerns about inclusion I'm fine with leaving it out.

We will be using this benchmark (and a few others we will probably add) extensively.

(Either way, it would still be nice to have a way to plug in custom benchmarks externally. See #89.)