Keep federated jobs until origin is up and synced

This allows the origin to be able to sync up jobs after it has been
down.