Support for multiple metric groups with multiple metric leaders

[cvonelm]

TL;DR: Add support for having multiple groups of metric events all with different metric leaders (and possibly metric readout intervals)

This is something I've come across in the perf documentation recently.

Using a syntax like -e {cpu_core/cycles, cpu/branch_misses} {cpu_atom/cycles, cpu_atom/mem_stores} it is possible to specify two metric readout groups with two different metric leaders in perf.

One use case is the #213 issue. As the topdown metrics require a custom metric leader (as far as I understand it), which is most probably only useful with topdown events and not "normal" events like cpu_core/cache_misses one would currently have to choose between either measuring topdown events or measuring any other event.

The biggest downside would be, that this might be a breaking change to the lo2s command line options.

[cvonelm]

I'm currently cleaning up the event reading code and I stumbled over this config check:

      if (arguments.provided("metric-count") && !arguments.provided("metric-leader"))
       {
           Log::fatal() << "--metric-count can only be used in conjunction with a --metric-leader";
           std::exit(EXIT_FAILURE);
       }
   
       if (arguments.provided("metric-frequency") && arguments.provided("metric-leader"))
       {
           Log::fatal() << "--metric-frequency can only be used with the default --metric-leader";
           std::exit(EXIT_FAILURE);
       }

Does anybody remember why we have these checks?