Consider support multi-containers pod error code handling

[hzxuzhonghu]

Currently, the LifecyclePolicy.ExitCode only takes the first container exit code of the pod.

So if users have multi-containers pod, it is impossible for them to specify the error handling behavior for each container in a pod.

I'd suggest to update the LifecyclePolicy API. Since ExitCode is just added, I think it is of little risk to do it now.

[hzxuzhonghu]

The initial thought

type LifecyclePolicy struct {
	ExitCode []ExitCode
}

type ExitCode struct {
      containerName string
      code *int32
}

[k82cn]

/kind feature
/sig apps

[TommyLike]

Do we care about any case of the exitCode or we just care about if there are any unsuccessful exitCode among the containers?

[hzxuzhonghu]

Honestly, in normal case i think it is not necessary. This is a advanced feature, maybe useful for those jobs that have some dependencies, for example a training job may return code x when the model data is unavailable, and return y when the data is of invalid format, return z when a temporary network connection err, etc.