Support for rbegin / rend on rmm::device_uvector

There is currently no support for rbegin / rend on rmm::device_uvector.

Currently, the solution is to use thrust::make_reverse_iterator.

I think supporting rbegin / rend would ease the porting of HPC CPU code that want to rely on rmm::device_uvector for memory. This would allow keeping the code that uses rbegin / rend as is.
