Cleanup inactive domains before running E2E tests

[dereknola]

We currently cleanup any running VMs at the start of the Drone E2E pipeline, to account for the case of a previous cancelled E2E Drone run. Recently, https://drone-pr.k3s.io/k3s-io/k3s/7228/6/4 was cancelled mid test. While the VMs were correctly destroyed next CI run, the now inactive domains were not, resulting in https://drone-pr.k3s.io/k3s-io/k3s/7251/6/4. We need to also make sure we cleanup the inactive domains during https://github.com/k3s-io/k3s/blob/master/.drone.yml#L754-L761.

[brandond]

We should probably also have the e2e pipeline depend on the amd64 pipeline, and not run if it fails. There is no point in running all the in-depth vagrant tests if we already know that the basic tests failed, and this would free up the runner for other tasks sooner.

Killing a run that someone knew was going to fail, partway through e2e, is probably how the vagrant domains got leaked in the first place.

[dereknola]

Since this is all CI related, with nothing to QA, I am pinning the backports to this issue as well.