Define who owns cluster add-ons (the actor updating the master or the cluster admin, assuming they are different) and what should happen to them when the master is upgraded:

Should we replace the DNS pods with new ones? Leave the old ones running and document for users how to update them?

We certainly don't want 2 sets of them running (or to ignore an error trying to start a second set if our intent is to update them).

Related work in #3579.

/cc @satnam6502