GKE Autopilot tests broken on GKE 1.26 about cloud-on-k8s

[pebrc]

/host/var/lib/kubelet# du -h /host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235
4.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/work/work
8.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/work
4.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/run/secrets/e2e
4.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/run/secrets/kubernetes.io/serviceaccount
8.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/run/secrets/kubernetes.io
16K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/run/secrets
20K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/run
...
722M	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/.cache/go-build
722M	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/.cache
44K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/tmp/artefacts
...
760M	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/tmp/go-build3966496234
761M	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/tmp
4.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/etc/e2e
8.0K	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs/etc
1.5G	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235/fs
1.5G	/host/var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/235

It looks like two things contribute to ephemeral storage:

1. the go-build artifacts in /tmp with 761M I don't think this is it because we mount a dedicated volume for that in autopilot e2e tests
2. the go build cache in .cache with with 722M which goes into the containers filesystem on autopilot as well but it is below the 1Gi limit

[pebrc]

I think the real problem is that the autopilot detection code does not work anymore in 1.26 which meant we were not actually mounting a dedicated volume to /tmp

// isAutopilotCluster will detect whether we are running within an autopilot cluster
// by using the `remotenodes` resource, which only seems to exist on autopilot clusters
// not standard GKE clusters.
func isAutopilotCluster(h *helper) bool {
	_, _, err := h.kubectl("get", "remotenodes")
	return err == nil
}

Verifying this now.