Representative matrix set identification for SpMV

[weiya711]

From @willow-ahrens: We need to consider adding a suite of N (maybe 20?) or so matrices that we consider "important" enough to warrant extensive study, as well as a few much larger suites of matrices to test interesting behaviors or properties. Since Spmv has been studied before, we can start with existing lists of benchmarks [5, 6, 7, 8] from #2.

Ideally, this would be used for the SpMV leaderboard (ranking of) submissions.

Pulling this task out from #2

[r0cketdyne]

Simply starting with existing lists of benchmarks may overlook specific criteria or considerations necessary for the intended use.