Test runs horrifically slow

[MattWindsor91]

Compare the run time of litmus7 on its defaults (equivalent to our -iterations=1000000 -period=100000) to the run time of phenolphthalein on the same: there's an order of magnitude slowdown.

This could be for a few reasons:

• The barriers we use might be overly slow; litmus's literature warns against the use of pthread barriers for this reason, for instance (perhaps use spinlocks or something else);
• The constant need for a thread to grab the mutex on the runner's shared state might also be slowing us down, so I'll consider possibly having the spawner thread perform this task instead.

I might see if I can profile this before making any changes.

[MattWindsor91]

Two notes:

• This is much faster in release mode, so debug mode must be seriously gumming up the atomic code;
• The spinner synchroniser seems to help, but it also suspiciously-infrequently produces weak behaviours, which I might file an issue over later.