Related to
#444

The rationale behind this is that we don't want to pollute worker job
queue with large file blobs. Instead, upon API request, we create a
pg_largeobject and use that in worker. After job is finished (with
error or not), the object is removed.