Add tests for large number of items & large number of segments to DeviceSegmentedSort

[elstehle]

#1787 has the goal to benchmark for different offset types. However, it is unclear, whether instantiating the Dispatch class template with these offset types will actually work correctly and whether the benchmark numbers are representative of properly working algorithm instantiations.

To make sure that all works as expected we want to add tests for large number of items and large number of segments for DeviceSegmentedSort.