Expand options of `scaling_benchmark`

[carstenbauer]

Similar to JuliaPerf/BandwidthBenchmark.jl#9 it would be great to expand the options of
scaling_benchmark(). Concretely, it would be nice to have the following options:

• return the individual bandwidths rather than just the maximum
• run only a single, specified kernel
• does it make sense to support running user-defined kernels (i.e. a function input argument)?