Set a default value for restartPolicy about training-operator

[paramgo]

Yes, not sure about the severity of the issue, however if you leave out the restartPolicy, it seems K8s 1.8 defaults to Always which results in the creation of master/workers but not ps (error below):

I  Creating Service: master-xyzz-0 
I  Service master-xyzz-0 already exists. 
I  Creating Job: master-xyzz-0 
I  master-xyzz-0 already exists. 
I  Creating Service: worker-xyzz-0 
I  Service worker-xyzz-0 already exists. 
I  Creating Job: worker-xyzz-0 
I  worker-xyzz-0 already exists. 
I  Creating Service: ps-xyzz-0 
I  Service ps-xyzz-0 already exists. 
I  Creating Job: ps-xyzz-0 
E  trainingJobCreateReplicas() error; [Creating Job ps-xyzz-0 returned error., Job.batch "ps-xyzz-0" is invalid: spec.template.spec.restartPolicy: Unsupported value: "Always": supported values: OnFailure, Never] 
  undefined
A simple TfJob configuration with one master, one worker, and one ps server would suffice where the restartPolicy for ps server is omitted (while it is present for master/worker).
from training-operator.

[ScorpioCPH]

After discuss, We will extend kubernetes built-in RestartPolicy by adding new policy ExitCode:

    RestartPolicyAlways    RestartPolicy = "Always"
    RestartPolicyOnFailure RestartPolicy = "OnFailure"
    RestartPolicyNever     RestartPolicy = "Never"
    RestartPolicyExitCode  RestartPolicy = "ExitCode"

We let users set this field according to their model code.

• If set RestartPolicy to OnFailure/Always, user should add reloading checkpoint code by themselves.
• Otherwise restarting will take no effect.

ExitCode policy means that user should add exit code by themselves, tf-operator will check these exit codes to determine the behavior when an error occurs:

• 1-127: permanent error, do not restart.
• 128-255: retryable error, will restart the pod.

[gaocegege]

dup with #524