Called by a compute node for a run's "Pennsieve Dataset" data target,
with the run's callback token. The target must be one of the run's
dataset targets, writing to dataset_id (one of the run's datasets);
the folder and conflict handling are the target's, from the run.
Registers a page of the target's staged files (imports/{runId}/{targetId}/
in the node's bucket) on its manifest and queues their copies into the
dataset's storage; the upload lambda then creates their packages.
Idempotent: the same run, target and dataset always use the same
manifest and upload ids. Max 1000 files per call, each up to 40 GiB.