refs: https://github.com/TryGhost/DevOps/issues/78
Instead of running a single instance of Ghost, we now run an instance of Ghost for each test worker.
This has the unfortunate effect that a test failing will close and restart a new instance of Ghost, but in general will be multiple times faster than sequential execution of tests.
- we should start to keep tests grouped by their area, so first we split
by Admin tests and then Portal tests, and within that we split into
setup/Tiers/Offers etc
