Currently, in many of our tests where it is possible, we start the VM in the fixture where we create the using a call to `vm.start()`, and afterwards we verify it completed its start using a later call to `running_vm()`.
This methodology saves time because it enables starting more than one VM in parallel, but there are occasions where it leads to a race, and the VM start ends with a failure.
We need to avoid starting the VMs with `start()` (except for specific places where it is required to do it like that), and instead start our VMs in an orderly manner using `running_vm()`.