Fix segmented-sort overlapped input/output indices

[davidwendt]

Description

Fixes call to CUB DeviceSegmentedSort::SortPairs where the input and output indices pointed to the same temp memory. The documentation from https://nvidia.github.io/cccl/cub/api/structcub_1_1DeviceSegmentedSort.html#id8 indicates the d_values_in and d_values_out memory must not overlap so using the same pointer for both created invalid output in certain conditions. The internal function was implemented to expect the input values to be updated in-place. The fix uses separate device memory for the input and output indices.

Closes #16455

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[davidwendt]

/merge