Add support for a kafka-powered replica of the graph.

This is in a dedicated compose file so that one must choose to run it (i.e.
it is not a service started by default when using a 'docker-compose up').

It:

• overrides the swh-storage service to activate the kafka direct writer on swh.journal.objects prefixed topics using thw swh.storage.master ID,
• overrides the swh-web service to make it use the replica instead of the master storage,
• starts a db for the replica,
• starts a storage service based on this db,
• starts a replayer service that runs the process that listen to kafka to keeps the replica in sync.