[FEA]: Integrate runtime-cooperative segmented reduce through internal dispatch

[anikaj-eng]

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.

The runtime-cooperative kernel and picker need to be reachable from DeviceSegmentedReduce through CUB's internal dispatch and tuning-policy path.

Describe the solution you'd like

Integrate the runtime-cooperative segmented-reduce path into internal dispatch so tuning policy can select it, with the existing CUB implementation as the fallback. Do not add a new public user-facing API.

Close once the path is selected through dispatch and covered by CUB tests.

Describe alternatives you've considered

No response

Additional context

Depends on the runtime-cooperative kernel and picker building blocks.

[anikaj-eng]

cc: @nvjonwong

[nvjonwong]

Looking into integration in the dispatch.