Add an other wait strategy (atomic flag with relaxed) for race_test

[martong]

This idea comes from crimson-virtue.
In ExecuteInLoop.hpp instead of relying on asm("") we could use an atomic ready flag with relaxed load and store.
In RandomWait::operator() we would have a while loop until the flag is not set. The flag would be set in the main thread. Because of cache synchronisation we might not know exactly when will each thread be notified about the change, this way they will wait for some random time.