Add --group flag to runtime benchmark/profiling command

[Kobzol]

This flag allows us to choose a specific benchmark group that will be compiled, when running runtime benchmarks or profiling. This serves purely as a latency reduction optimization for local experiments, because now even with --no-isolate it takes some time until a runtime benchmark starts executing, because all compiled groups are currently executed first with the list parameter. And since some of the groups perform non-trivial work in main (e.g. compression currently), this can take a few seconds, which is annoying.