exchange: improve computation of relevant markers for large repos Find the candidates for nodes with relevant markers by going over all markers instead of iterating over all nodes. Most nodes will not have markers anyway. Further optimize the code by allowing revsets as well, which reduces the materialization cost.