Replaced explicit algorithms for host in VectorOperations with general implementation

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.