This allows int-vectors to be multiplied by float-scalars, producing
accurate int values (e.g. 4.2 * 10 == 42).

According to benchmarks, there is practically no difference in
performance. Only explicit unrolling is helpful, but that has been
implemented for the general algorithm in Reduction::reduce as well.

- separate data types for input arrays of binary operations
  (necessary for the equality comparison of two arrays of different
  type)
- added separate template parameter for ResultType, which defaults to
  the RealType
  (the goal is to allow sum of bool vectors into int instead of bool)

This prevented scaling of GMRES on small grids (up to 240x240)!

Implemeting heat equation benchmark.

RowLengths was renamed to CompressedRowsLengths.
OpenMP support wads added to matrix-vector multiplication and to scalar
products.
CSR format can use Cusparse for SpMV.