The main motivation is to move toward running those tests for a specific
target, that is, if a test won't run on Windows, then we shouldn't build
it. This was previously the default behavior and prevented a successful
run on AppVeyor.
I borrowed this pattern from the tests in the Cargo project.
