[autoscaler][kubernetes] autoscaling hotfix

[DmitriGekhtman]

Why are these changes needed?

Right now Kubernetes fill_out_available_node_type_resources logic fills "GPU":0 for non-gpu nodes, which interacts badly with the resource demand scheduler's gpu conservation logic, preventing autoscaling on k8s.
This fixes KubernetesNodeProvider's resource-filling logic to not fill fields with value 0.

Related issue number

Checks

• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

Did quick manual check that this fixes the problem.
Will add more test logic to the K8s operator unit test (not currently in CI) later.