Support trace-children

[osiewicz]

https://valgrind.org/docs/manual/manual-core.html
--trace-children allows one to profile processes spawned by the profiled process. This should come in handy for profiling things like miri, which are spawned by other processes.

Note that this will require some changes:

• It will no longer be guaranteed that a single benchmark has only a single output profile (that'll be a breaking API change).
  • How do we sort the output list of profiles?
• Currently, we parse only a single file based on a PID of a spawned process or an output file flag. Given that we might have multiple output files, the benchmarks should be executed within a tmp directory and then - if user wants to preserve the results - moved into cwd/output file path on demand. This way, we'll know for sure that the contents of a given temp dir contain only results of single benchmark run.