Add mode to persist driver between test runs to avoid recreating filesystem

Currently, a new driver is created for each individual test as a way to run tests in isolation. This is fine for the docker driver, but for the tar driver this requires us to recreate the filesystem for every test, which can take on the order of half a second. Since the tar driver is incapable of running commands anyway, tests can't interfere with each other, so we should have a way to persist the driver between tests.