Support Parallel save result with export_workspace.

[HansVRP]

Example Job: j-25011314204947baa2057fa6f64bae8b

More info: spark is perfectly fine with running multiple jobs and stages concurrently. By doing so, we can keep executor allocation rate high.
So in the presence of multiple 'save_result' nodes, this could really help to improve overall performance.

[JeroenVerstraelen]

TODO: Estimate the amount of work this issue will require

[JeroenVerstraelen]

LCFM also benefits from this issue

[EmileSonneveld]

A quick attempt to distribute the SaveResult nodes to multiple python threads gave an error when trying to serialize the RDDs inside it.
RuntimeError: It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transformation. SparkContext can only be used on the driver, not in code that it run on workers. For more information, see SPARK-5063.

_pickle.PicklingError: Could not pickle the task to send it to the workers.

It might be needed to run the ProcessGraphDeserializer.evaluate for each thread too.
Or find a way to start the write_assets tasks all a once from the main thread, and then start polling for the results also on the main thread.

[JeroenVerstraelen]

Would something like collectAsync help in this case:

val value = rdd.collect() //RDD elements will be copied to spark driver 
val value = rdd.collectAsync() //no copy here  
value.get() //Now, RDD elements will be copied to spark driver