Pynumero.sparse transpose

[michaelbynum]

We need to revisit BlockMatrix and MPIBlockMatrix transpose methods. Currently, they require copying because supporting a transpose without copying is difficult with nested blocks.