[FEA]: Support fancy iterators in cuda.parallel

[gevtushenko]

Is this a duplicate?

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

Area

General CCCL

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

Compared to existing solutions in PyTorch and CuPy, one of the distinguishing features of cuda.parallel is flexibility. Part of that flexibility is coming from support of user-defined data types and operators. But compared to CUDA C++ solution, cuda.parallel API is still limited. We are still missing fancy iterators and cache modified iterators in cuda.parallel.

Describe the solution you'd like

Given that fancy iterators support might require rather invasive changes to cuda.parallel and CCCL/c libraries, we should design fancy iterators before introducing more algorithms to cuda.parallel.

Describe alternatives you've considered

No response

Additional context

No response