[FEA]: Use LRB-based segmented sort as default untunable backend #12541

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

CUB

Is your feature request related to a problem? Please describe.

General tuning interface for segmented sort is a large effort. While we are working on it, users of segmented sort miss 5-10x speedup we could've provided today.

Describe the solution you'd like

This issue can be closed when cub::DeviceSegmentedSort:

• uses LRB-based implementation by default in cases when a. it's faster, and b. user doesn't ask for tuning
• dispatch to legacy implementation for source compatibility when user provides custom tuning

Describe alternatives you've considered

No response

Additional context

No response