Currently, we cannot stream the input stream directly from s3 into the stdin of the worker. This is because currently the inbox->processing->outbox flow (needed by the workstealer) is not separate from the data to be processed. We will need to keep them separate, somehow. So that the inbox->processing->outbox flow is done with markers, and the underlying data is placed in a separate folder. This will also avoid potentially expensive copying of large data blobs through those stages.