Looking at:

https://testing.whamcloud.com/test_sets/a9c1ab0e-2d95-11e9-b3df-52540065bddc

We see that when test_93 fails, test_98 also fails because the system is not in the expected state.  Most tests in conf-sanity assume the state is "cleaned up" when they start, but most of them also don't cleanup correctly on errors.

Adding cleanup to CLEANUP so it's run after every test, guaranteed, can fix most of this.