This could be as simple as providing find_subgraph as a default for EmbeddingComposite's find_embedding parameter, but at the same time replacing expensive BQM embedding / sampleset unembedding with simpler variable relabeling will result in significant performance boost.
See also #579.
