Propagate flexvolumes errors

[LilliJane]

Expected Behavior

Users should be able to identify a flexvolume related issue easily.

Actual Behavior

At the moment we don't have any easy way to inform our users about what's happening with flexvolumes (from expected behavior to important errors).

One option could be to store the metadata in the dataset (#255).

This issue is opened so we can share ideas and opinions about a new design

[LilliJane]

If an output dataset is too big, the flexvolume will try to upload it, which will fail. In this case, we just delete the dataset and unmount the volumes, but users don't have any way to understand why their output dataset is empty.

[LilliJane]

It would be interesting to publish events to propagate the errors but also the status of a dataset. And with Kubernetes 1.9, they introduced a way to filter events by objects.