[k8s] sky down --purge does not work for k8s when switched to a new cluster

[romilbhardwaj]

sky down --purge for k8s stays stuck on Terminating 1 cluster when a new k8s cluster is connected. Looks like a recent regression, since it works fine on 0.4.1.

Some initial digging indicates this call to ray stop on the new provisioner gets stuck:

skypilot/sky/backends/cloud_vm_ray_backend.py

Lines 3828 to 3836 in cf49028

	# Stop the ray autoscaler first to avoid the head node trying to
	# re-launch the worker nodes, during the termination of the
	# cluster.
	try:
	# We do not check the return code, since Ray returns
	# non-zero return code when calling Ray stop,
	# even when the command was executed successfully.
	self.run_on_head(handle, 'ray stop --force')
	except exceptions.FetchIPError:

Related to #3013.

Repro:

1. Create GKE cluster
2. sky launch -c test
3. sky local up to switch k8s cluster identity
4. sky down test --purge stays stuck.

[romilbhardwaj]

Raised by user - should fix this asap.

[romilbhardwaj]

Closed by #3043.