Add Intrinsics for 64 Bit Atomics

[tgymnich]

https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf

64 bit atomics are supported on M3 and up with some support on M2.
For this to be usable we need to figure out how to guard this feature to M2 or M3 and up by checking for the MetalGPUFamily.

[maleadt]

For this to be usable we need to figure out how to guard this feature to M2 or M3 and up by checking for the MetalGPUFamily.

That's pretty hard, for kernel-level functionality. We could expose it similarly to the Metal and AIR version, but then all we can do is emit an exception (and thus a run-time error, which are currently still invisible until #416 is finished). Maybe it'd be worth trying to pick those up in the GPUCompiler validator? Seems hacky; in Julia those things are typically postponed until run-time.

[tgymnich]

We could also check in the compiler for the intrinsics and report back. Just like @printf does it.

[maleadt]

We could also check in the compiler for the intrinsics and report back.

Slowly making Julia a static language 😛 I agree it's better for the user, though.