feature ask: support for loops

[SinaChavoshi]

could we add loop operations to _ops_group.py

There is strong need for ability to run steps within a pipeline in a loop, for example train on ### epochs if accuracy is below X% rerun the training.

current work arounds are very cumbersome and messy:

1. Schedule job at start up check for last run status
2. Create a massive graph with early stopping
3. Manage the loop out side of the pipeline

[gaoning777]

The recursion is supported now. https://github.com/kubeflow/pipelines/blob/master/samples/basic/recursion.py