Add microbenchmark

[stgeke]

So far MPI benchmark packages do not assess the performance of sparse collectives. Possible benchmarks need to include:

• different message sizes and virtual topologies
• blocking and non-blocking performance