Fix setting defaults.

[jlewi]

• Need to register the default functions.
• In setup we need to invoke setting the defaults on the objects.
• This fixes a break introduced when we refactored the code.
• Fix fail to run example job. invalid job spec: tfReplicaSpec.TfPort can''t be nil #284

---

This change is 

[coveralls]

Coverage increased (+0.8%) to 31.43% when pulling f087e23 on jlewi:fix_defaults into e619ac4 on tensorflow:master.

[gaocegege]

The job in prow is weird: https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/pr-logs/pull/tensorflow_k8s/299/tf-k8s-presubmit/354/

Result | SUCCESS
-- | --
Tests | 5 failed / 5 succeeded

[wackxu]

@jlewi Sorry for the mistake and Thanks for your fix. Can we keep style with k8s and add function below:

func init() {
	// We only register manually written functions here. The registration of the
	// generated functions takes place in the generated files. The separation
	// makes the code compile even when the generated files are missing.
	SchemeBuilder.Register(addDefaultingFuncs)
}

[jlewi]

@wackxu NP. What file is the init function in?

[wackxu]

@jlewi The one you modifed pkg/apis/tensorflow/v1alpha1/register.go

[jlewi]

@wackxu PTAL

[jlewi]

The job in prow is weird: https://k8s-gubernator.appspot.com/build/kubernetes-jenkins/pr-logs/pull/tensorflow_k8s/299/tf-k8s-presubmit/354/

#280 (comment)

/cc @gaocegege

[wackxu]

/lgtm

[coveralls]

Coverage increased (+1.0%) to 31.609% when pulling b51222a on jlewi:fix_defaults into e619ac4 on tensorflow:master.

[wackxu]

@jlewi Seems there have an gometalinter error, just run goimports -w pkg/trainer/training.go

pkg/trainer/training.go:1::warning: file is not goimported (goimports)

[gaocegege]

Should we add addKnownTypes here:

SchemeBuilder.Register(addKnownTypes)

And here https://github.com/tensorflow/k8s/pull/299/files#diff-2f7eaaeac6f6da3e4c93e8530a8a4e4aL10 should be:

SchemeBuilder      runtime.SchemeBuilder

I think it may be clearer.

[jlewi]

@wackxu ?

[wackxu]

Do not think so, I think now is better and there is nothing that are confuse.

[gaocegege]

Alright, then LGTM

[coveralls]

Coverage increased (+1.0%) to 31.609% when pulling beee7f8 on jlewi:fix_defaults into e619ac4 on tensorflow:master.

[coveralls]

Coverage increased (+1.09%) to 31.714% when pulling beee7f8 on jlewi:fix_defaults into e619ac4 on tensorflow:master.

[gaocegege]

And this PR also fixes #297