DinD tracking issue

[BenTheElder]

We use Docker in Docker pretty successfully for various things in test-infra but it needs some work.

• 1. We need an image with DinD and podutils support to supersede images/bootstrap, see also figure out podutils and init / zombies #8779, figure out DinD + podutils #8780

  • we can probably just refactor the ENTRYPOINT runner a bit to allow it also being used by certain jobs, without the bootstrap bit and with the real test command instead.

• 2. We should make our existing support a pod-preset so that it's easier to use. At this point it's used widely enough that we should remove all of the copy paste for env / volumes. This will make it easier to refactor how the support works.

• 3. We have a couple of areas where we've worked on running "local clusters" within the test as a cheaper / more hermetic way to test, but none of these is quite fully baked, and they have a few issues integrating currently (mostly improper cleanup / making sure they work with the DinD they are already on from 1) when doing DinD on on [dindind etc...], but some others as well).. We should improve these. To that end I am in particular taking another look at dind/.

/area jobs
/assign

[BenTheElder]

#3 moved to sigs.k8s.io/kind. will migrate tracking of remaining work to there generally.