To make sure corruptions such as T680 don't happen again, use the same
normalization function as swh.model before inserting timestamps into our
database.

This makes swh.storage reject non-integer timestamp values as well.

Update tests to reflect this change.

Close T672

The buckets are using the last two bytes of the object id, so that we spread the
load across different lines on sequential archivings.

Closes T646

The check_config method allows a dynamic check of the configuration for
a running storage. We can make sure that we have proper permissions on
the object storage as well as the database before running things.

Related T575

The 'unknown sha1 path' cannot happen in the default archiver since it
reads from the archive db (so the fallback code is not necessary in the
worker).

To the contrary, since 'archiver to backend' reads from stdin (for now),
we could have unregistered sha1s from that source.

This commit makes the director deal with that before sending sha1 to
workers.

It's also the director's job to set the state to 'missing' when the
force_copy is true before sending sha1 to worker.

Before that, it could for example pushed copies to other mirrors where
the content was missing.

Closes T569

If you want the archiver to have cloud abilities, install the
package python3-swh.objstorage.cloud on the server as well.

Logging an entry about it

- Only read the storage key once.
- Improve the logging error.

Also, adds a force_copy flag in the configuration file to avoid checking
preexistence of sha1.

This is to be efficient for the first time copy in a new backend.

The initial director and worker had split configuration files.
Now it's unified to be both archiver/worker.yml file