Add nested loop kernel reduction benchmark

[johnbowen42]

This topic came up during a discussion in the RAJA dev meeting. RAJAPerf currently doesn't measure and compare the performance of a nested kernel reduction benchmark to native CUDA/HIP/OpenMP. Adding this as a benchmark to RAJAPerf will be especially useful for evaluating the performance of LLNL/RAJA#1799.