[k8s] Ray cluster is incorrectly initialized with gpu resource

[romilbhardwaj]

On a CPU-only kubernetes cluster, the ray cluster shows 1 GPU available, which is incorrect. Ssh into node and run ray status:

(base) sky@test-2ea4-ray-head:~$ ray status
======== Autoscaler status: 2023-09-25 23:08:16.304098 ========
Node status
---------------------------------------------------------------
Healthy:
 1 ray_head_default
Pending:
 (no pending nodes)
Recent failures:
 (no failures)

Resources
---------------------------------------------------------------
Usage:
 0.0/2.0 CPU
 0.0/1.0 GPU
 0B/2.00GiB memory
 0B/476.84MiB object_store_memory

Demands:
 (no resource demands)

Repro:

1. sky local up - this creates a CPU only cluster
2. sky launch -c test --cloud kubernetes -y
3. ssh test
4. ray status