Add a command to profile a runtime benchmark

[Kobzol]

This PR adds a CLI command for profiling a single runtime benchmark. Currently, running a single benchmark function under cachegrind is supported. Later, I want to make the cachegrind profiling more precise with cachegrind client requests that would only profile the part of the benchmark group where the benchmark function actually runs (so without initialization, CLI parsing,e tc.). I also want to add cachegrind diff functionality in a followup PR to compare instructions executed by a runtime benchmark compiled by two different compiler versions.

I named the command just profile_runtime instead of profile_local_runtime. Profiling of "remote" artifacts doesn't really make sense, so I think that profile_local should also be renamed to profile_compile or something like that.