Fix for perf_event multiplexing

Created by: TomTheBear

In most cases it is not required to do multiplexing because LIKWID never uses more events than there are physical counter registers for CPUs. The Intel Skylake SP architecture (and maybe Cascadelake SP) has a bug in the PMC3 if transactional memory is enabled (see #224 or likwid-users mailing list post). This causes that perf_event multiplexes on the remaining 3 counters and LIKWID has to evaluate the multiplexing information to calculate the real counter value.