Hypersparse matrices seem quite performant and come with the benefit of *not requiring relabelling*. We should give them a try. https://github.com/szarnyasg/LAGraph/tree/hypersparse-test