The internal Thrust sequential system should be replaced by libcu++

Thrust maintains a large implementation of [sequential algorithms](https://github.com/bernhardmgruber/cccl/blob/3314a70aa01e477c8f645da810e9810002f80483/thrust/thrust/system/detail/sequential). These are used as fallbacks when Thrust algorithms on the CUDA system are called from device code (CDP). The entire backend could, and should be replaced by the libcu++ algorithm implementation. This reduces Thrust's code base and maintenance effort, while increasing test coverage of libcu++.