[k8s] Custom GPU node labels need better error hints

[romilbhardwaj]

A user configured their k8s cluster GPU labels in uppercase:

kubectl label nodes <node> skypilot.co/accelerator=A40

sky check passed silently:

$ sky check
...
Kubernetes: enabled
...

But sky launch fails:

(base) ➜  ~ sky launch -c test --gpus A40:1
I 10-03 17:39:46 optimizer.py:1044] No resource satisfying <Cloud>({'A40': 1}) on [AWS, Azure, GCP, IBM, Kubernetes, Lambda].
sky.exceptions.ResourcesUnavailableError: Catalog and kubernetes cluster does not contain any instances satisfying the request:
Task(run=<empty>)
  resources: <Cloud>({'A40': 1}).

To fix: relax or change the resource requirements.

Hint: sky show-gpus to list available accelerators.
      sky check to check the enabled clouds.

Neither of these hints are useful (sky show-gpus will be supported only after #2638, and sky check showed no error).

The problem is the user used uppercase labels to label their nodes, and our checks only check the presence of a skypilot.co/accelerator label, not its validity. Our hints should include this.