[Perf] Improve perf-test YAMLs and README

[kevin85421]

Why are these changes needed?

TODO:

• Prometheus metrics
• If RayJob custom resources cannot complete indefinitely due to the lack of gang scheduling, it appears that the clusterloader process does not time out even after 20 minutes.

Related issue number

2102

Checks

• [ ] I've made sure the tests are passing.
• Testing Strategy
  • [ ] Unit tests
  • [ ] Manual tests
  • [ ] This PR is not tested :(

I modified the script a bit to create 10 RayJob CRs on my local Kind cluster.

[kevin85421]

cc @andrewsykim