Parallelize reads - #1607

@lossyrob Grisha has opted here not to explicitly use Tasks with a custom thread pool. Apparently nondeterminism handles the specifics of that.

He was seeing similar issues to what we had before, where Tasks weren't appearing to complete and using pool.shutdown() was causing problems. We might want to consider back-porting his solution here to our other work.

Without dedicated pool it will end up using scala.concurrent.ExecutionContext.Implicits.global which is comes with this info:

The implicit global ExecutionContext. Import global when you want to provide the global ExecutionContext implicitly.

The default ExecutionContext implementation is backed by a work-stealing thread pool. By default, the thread pool uses a target number of worker threads equal to the number of available processors.

I think that works out alright for spark tasks

@echeipesh in terms of scalaz, it would use DefaultStrategy, which is FixedThread pool with threads = available processors amount. Thx for a pointer, though I am still curious (as an example of what i am talking about S3RDDWriter), creating a batch of tasks using our custom thread pool, how tasks would be scheduled with nondeterminism.njoin as it uses DefaultStrategy and DefaultExecutor?

def njoin[A](maxOpen: Int, maxQueued: Int)(source: Process[Task, Process[Task, A]])(implicit S: Strategy): Process[Task, A]

This continues to be an informative PR.