Port non-standard `thrust` algorithms to `cuda`

We want to move thrust parallel algorithms that are not eqivalent to standard algorithms and move them into `cuda`

This has the clear benefit that we can rely on a single namespace and can modernize the existing infrastructure