Tpetra::Crs{Graph,Matrix}: Thread-parallelize row offset computation
Created by: mhoemmen
@trilinos/tpetra CrsGraph and CrsMatrix currently compute row offsets from row counts sequentially. We need to thread-parallelize this. Use Kokkos::parallel_scan. See the ComputeRowOffsets functor in Tpetra_Details_FixedHashTable_def.hpp (we should actually add a computeRowOffsets function that uses this functor, and promote that function to its own header file).