[Subtask][operator] svc generation refinement

[advancedxy]

Code of Conduct

• I agree to follow this project's Code of Conduct

Search before asking

• I have searched in the issues and found no similar issues.

Describe the subtask

We should reconsider the service generated by RSS operator as it may not make too much sense.
Current behavior for RSS to generate svcs:

1. generate NodePort service for coordinator, which is mandatory
   • the svc is passed to shuffle server as coordinator address
   • it requires RSS configuration must defines RPCNodePort and HTTPNodePort, which is restrictive and cost node ports
2. NodePort svc is generated for shuffle server if node port is supplied
3. headless svc is generated for shuffle server.

There are some ways to improves this situation:

1. since coordinator and shuffle server are both in the k8s cluster, there's no need to use ClusterIP/NodePort for shuffle server to connect with coordinator. Therefore headless svcs should be created for coordinators, and passed to shuffle server
2. it's pointless to generate svc for shuffle server as it should be accessed directly per node.
3. for inter-cluster only workload, there's no need for coordinator to expose NodePort. Therefore NodePort configuration in RSS should be optional rather mandatory.
4. to expose RSS cluster deployed on K8S to external compute resource, such as a yarn cluster, then coordinator and shuffle server should both be deployed with hostnetwork. The service part is preferred with LoadBalance.

Parent issue

#462

Are you willing to submit PR?

• Yes I am willing to submit a PR!

[advancedxy]

cc @wangao1236