Change segmented scan tests to use cuda::buffer

I'd recommend using `cuda::buffer` instead. `thrust::device_vector` is quite slow for compile time.

_Originally posted by @jrhemstad in https://github.com/NVIDIA/cccl/pull/10855#discussion_r3798818164_

Locate the segmented scan tests and identify their uses of thrust::device_vector; read the surrounding test setup before making the change. Replace the test buffer usage with cuda::buffer, then run the segmented scan test target and confirm the tests still pass.

Written by the indexing model from the issue text.