Investigate replacing little-lock exclusives with mutex_arc in taskgroup code.

[bblum]

I don't at all expect this change to go through, and would be surprised if it were nearly as fast, and also it requires moving sync and arc to core... but I'd like to see the difference to get a grasp on how they perform.

For a rainy day.

[bblum]

So investigated! Not only did it give a performance hit (like 10%, just by eyeballing the numbers, on task-perf-jargon-metal-smoke), as expected, it also had the interesting side effect that task-perf-linked-failure did not work properly because the mutex_arc supports killing tasks in the middle of it and poisoning the arc.

tl;dr: not just slow, also broken for an obtuse reason.