Now that we know that LLP works well to improve graph compression (25% size reduction in @seirl experiment on the full Merkle DAG), we should add it to the default graph compression pipeline implemented by swh graph compress.