This will allow us to have CI running concurrently on multiple PRs,
since we get 30 parallel jobs on Azure Pipelines but only 1 on AppVeyor.
We have parameterized --use-venv since AppVeyor was using it, but Azure
Pipelines was previously not.
Profiling on Azure Pipelines indicates that the majority of our time is
spent waiting for filesystem operations to complete. As a quick way to
improve our test speed in this area, we can do operations on a RAM disk
instead of the default SSDs.
