scan clause for `do concurrent reduce`

[brandongc]

OpenMP has a scan clause for reductions.

There is some nontrivial design space in good concurrent implementations of this primitive:

• http://www.eecs.umich.edu/courses/eecs570/hw/parprefix.pdf
• https://software.intel.com/content/www/us/en/develop/articles/openmp-simd-for-inclusiveexclusive-scans.html

[certik]

Indeed, I have heard from multiple people that we should consider adding it. Kokkos has it too: https://github.com/kokkos/kokkos/wiki/Kokkos::parallel_scan