[k8s] Handle kubernetes config and reachability errors

[landscapepainter]

This resolves #3013.

When ~/.kube/config was removed externally, two errors were noted each from running sky launch and sky down --purge.

sky launch: removing ~/.kube/config sets the Kubernetes cloud to be disabled. And as it was removed externally, the change does not get reflected to the state. To resolve this issue, the change was made to make sure sky check is ran when provisioning to update the state, so that write_cluster_config does not try to upload the credential file for Kubernetes cloud(~/.kube/config).

sky down --purge: runs cleanup_ports as part of the post_teardown_cleanup. While doing so for Kubernetes cloud, it tries to load the config and raises an error since it's removed. Error handling is implemented to handle the scenario.

Tested (run the relevant ones):

• Code formatting: bash format.sh
• Manual tests for two issues mentioned in [k8s] sky down --purge fails on Kubernetes clusters that no longer exist #3013
• All smoke tests: pytest tests/test_smoke.py
• Relevant individual smoke tests: pytest tests/test_smoke.py::test_fill_in_the_name
• Backward compatibility tests: bash tests/backward_comaptibility_tests.sh

[romilbhardwaj]

Thanks @landscapepainter!

[romilbhardwaj]

To fix the whole user story ("sky down --purge doesn't work"), can this PR also fix #3093?

[romilbhardwaj]

Now also closes #3093.

Since the missing kubeconfig is actually an error, I have removed the silent disabling of Kubernetes cloud (cloud-specific sky.check()) and instead added a note in the error message for the user to run sky check if they want to disable Kubernetes.

Tested against repros from #3093 and #3013.

[Michaelvll]

Thanks for fixing this issue @landscapepainter!

[romilbhardwaj]

Thanks @Michaelvll - addressed your comments.

[Michaelvll]

Thanks for the fix @landscapepainter @romilbhardwaj! LGTM.