Tpetra: Import/Export speed

[tjfulle]

@rppawlo has shown that, in some problems, Tpetra's Import/Export operation can take a significant amount of time - as much as 98% of a computational task's time. @jhux2 and @tjfulle isolated the slow down to a single loop copyAndPermuteStaticGraph. We refactored that section of code to treat the special case that both the source and target matrices were locally indexed here. There were modest speed ups of about 10%.

To achieve maximum speed up, @jhux2 and I think this section of code could be refactored to use Kokkos hierarchical parallelism along with replace*Values.