Copy finished data to secondary scratch

[soichih]

(Consider)

We create a copy of finished data when child task tries to process the data. But we don't proactively create a copy in case the resource created the data goes off line. This happens more often than not.

Another use case to consider is when a cluster has a very limited scratch space and data that task points to is removed much sooner than we'd like.

When a job finishes, we could automatically rsync the content to configured secondary storage (such as Wrangler).

Wrangler