ci(tofu-apply): Run job from WMCS runners
When we get to the point of trying to talk to the deployment-prep hosted k8s cluster we will need to be on the internal Cloud VPS network. We could try to rig up an ssh tunnel as described in https://phabricator.wikimedia.org/T372498#10065582, but it should also work just to run the job from inside the network in the first place. If this turns out to be flaky somehow we should look into setting up a custom runner just for deployment-prep.
Logically a revert of the revert at dcfc058d with minor differences for intervening changes.