[graceli24]

• Small modification to avoid work in initializing min_incident_edge_map from the edge_endpoints_array
• Use a local cache of min_incident_edge_map collecting to minimize number of async visits. This was a bottleneck for larger numbers of compute nodes since many async visits were made to the same supernode reps.
• Add draft implementation of writing point ids and cluster labels to a single-rank Metall datastore. This implementation is currently slow.
• Some general code clean up, including adding [[maybe_unused]] to lambda.