[PATCH] Atomic-IR pass: move out into generic area

[Tim Northover]

>   I don't see anything terrible in the pass or how you've written it. I can't comment on the atomic lowerings itself without a much closer look.

Thanks very much for taking a look, Eric. I'm reasonably confident
about the lowerings themselves (far more than I am about even the
existing MachineInstr stuff; I spotted a couple of dodgy points while
porting, around extension & max/min).

I've committed it as r206485.

Tim.