[k8s] sky check detects unlabeled nodes

[SeungjinYang]

Adds a hint to kubernetes sky check for any nodes with accelerators that are not properly labeled for use with SkyPilot.

The logic to detect contexts without proper labels can later be used to automatically label GPUs on those contexts.

Related issue: #4958
Fixes #5023

Tested (run the relevant ones):

• Code formatting: install pre-commit (auto-check on commit) or bash format.sh
• Manually run sky check kubernetes on a cluster w/o labeling it beforehand
• Manually run gpu label formatter script to check that jobs are only launched on unlabeled nodes.

$ python -m sky.utils.kubernetes.gpu_labeler   
Found 2 unlabeled GPU nodes in the cluster
Using default RuntimeClass for GPU labeling.
Created GPU labeler job for node ip-192-168-14-191.ec2.internal
Created GPU labeler job for node ip-192-168-24-7.ec2.internal
GPU labeling started - this may take 10 min or more to complete.
To check the status of GPU labeling jobs, run `kubectl get jobs -n kube-system -l job=sky-gpu-labeler`
You can check if nodes have been labeled by running `kubectl describe nodes` and looking for labels of the format `skypilot.co/accelerator: <gpu_name>`. 
$ ... # after some time
$ kubectl describe nodes | grep "skypilot.co"
                    skypilot.co/accelerator=v100
                    skypilot.co/accelerator=t4
$ python -m sky.utils.kubernetes.gpu_labeler 
No unlabeled GPU nodes found in the cluster. If you have unlabeled GPU nodes, please ensure that they have the resource `nvidia.com/gpu: <number of GPUs>` in their capacity.

[Michaelvll]

Awesome @SeungjinYang!

[Michaelvll]

Thanks @SeungjinYang! LGTM.