Arvados

I think it is desirable that we create a separate bug for each of the client libraries. This helps with reviewing, testing. That way we can also divvy up the work more effectively.

I think the general approach is:

• Add a num_retries argument to Arvados::Request::process_request. It handles the actual retry loop (a la Python), backing off between requests that should be retried.
• Add num_retries arguments to Arvados::ResourceMethod::execute and the save, update_attributes, and reload methods of Arvados::ResourceProxy. These should simply be propagated down to the appropriate lower calls, so they reach Arvados::Request. This should be a "keyword argument" in the methods that already take those.

It's not expressly specified but I figure this story wants to see crunch-job use this support.