The actual dependency to the current code does not depend on the
latest yet.

The retrieval of archives can temporarily build an archive from
multiple ones before sending its content to the client.  As a result,
when retrieving a content, we leave in a temporary location an
archive.  This temporary archive is referenced in a table
deposit_temporary_archive.

This adds a service which is in charge to poll regularly and delete
old temporary archives.

This possibly aggregates multiple archives into one and stream the
result to the client.

This is incomplete for now as the possibly temporary archive
referenced is left on the server.

Fix missing runtime dependencies on swh-*.

WIP, this still needs:
- integration testing (manual so far)
- clean up routine
- scheduler api interface (for now this uses celery directly)

- Fix error in PUT EM_IRI implementation
- Improve test docstrings
- Separate delete tests from update ones

It's already declared at the common swh.docs dependency