* change ksonnet install export version
* Upload TFJOB Using GPUs example output & Job Scheduling API verison
* Upload TFJOB Using GPUs example output
Change --batch_size diff.
* Update tftraining.md
1. `ks generate tf-job-simple-v1beta1 ${CNN_JOB_NAME} --name=${CNN_JOB_NAME}` will cause error:
```
ERROR handle object: patching object from cluster: merging object with existing state: unable to recognize "/tmp/ksonnet-mergepatch596037217": no matches for kind "TFJob" in version "kubeflow.org/v1beta1"
```
And tfjobs that created is with namespace 'kubeflow'
* Update content/docs/components/tftraining.md
* In progress: Refactoring the left-hand navigation panel.
* Added .description to section index pages.
* Removed the redundant Components Overview as the new index page of the components section does a better job.
* Added redirects for the files I've moved/deleted.
* Fixed the URLs in all doc links.