[FEA]: [CUB]: support passing in no initial values for various reduce + scan algorithms

### Is this a duplicate?

- [x] I confirmed there appear to be no [duplicate issues](https://github.com/NVIDIA/cccl/issues) for this request and that I agree to the [Code of Conduct](CODE_OF_CONDUCT.md)

### Area

CUB

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

#9289 exposed the no initial value option for `cub::DeviceReduce` for `run_to_run` determinism. We should extend this to other algorithms that currently expect an initial value:

- [ ] `cub::DeviceReduce`: `not_guaranteed` + `gpu_to_gpu` determinism
- [ ] `cub::DeviceScan`: `ExclusiveScan` always expects an initial value, while `InclusiveScan` does not, but has a separate `InclusiveScanInit` overload. We should deprecate this extra overload and use a similar solution to `DeviceReduce`. We should also make sure to support all determinism levels.
- [x] `cub::DeviceSegmentedReduce`
- [ ] `cub::DeviceSegmentedScan`
- [ ] `cub::DeviceScan::ExclusiveScanByKey`

### Describe the solution you'd like

An overload that allows the caller to omit the initial value for the above algorithms, without having a separate function like `InclusiveScanInit`.

### Describe alternatives you've considered

_No response_

### Additional context

_No response_

Start with the existing no-initial-value cub::DeviceReduce behavior introduced in #9289, then compare the remaining listed reduce and scan algorithms. Define consistent overload and determinism behavior, including the InclusiveScanInit transition; the work is done when the unchecked algorithms support omitted initial values without separate function names.

Written by the indexing model from the issue text.