Massage CityHash to work at comptime

[SpexGuy]

@sinitax ran into issues trying to use CityHash at compile time. This PR modifies the implementation and tests to work at compile time.

[jedisct1]

Looking good. Does it have any performance implications?

[SpexGuy]

Performance in debug mode may be impacted by the new offsetPtr function. We could mark offsetPtr and fetch32/fetch64 as inline to mitigate this. But I don't think I've introduced any new bounds checks. Runtime performance in optimized mode should be the same. The comptime execution for the tests is pretty slow (~15 seconds on my computer), so that will affect the total std test time. The change to the tests to write individual bytes instead of type punning may also cause a small increase in test time.

[LemonBoy]

What about

returnmem.readIntNative(u32, offsetPtr(ptr, offset)[0..4]);
// orreturnmem.readIntNative(u32, ptr[offset..offset+4]);

Ditto for fetch64 (and the other two identical fetch helpers? Can you just pull them at top level and delete a copy?)

[SpexGuy]

Huh, I thought these wouldn't work at comptime but apparently they do. Since the next line is byte swap, I can replace the whole function with readIntLittle.

[offset .. offset + 4] works at comptime but at runtime this would produce a slice, so I'm sticking with the other one.

[ghost]

This should work at runtime:

returnmem.readIntLittle(u32, ptr[offset..offset+4][0..4]);

(With slices, you can do [offset..][0..4] which is a nice little idiom, but that doesn't work for length-less pointers)

[SpexGuy]

That works but introduces a new but useless runtime check in debug modes (useless because it's only checking that offset + 4 - offset == 4).
ptr[offset .. offset + 4].ptr[0..4] would work, but for consistency I think it's worth sticking with offsetPtr everywhere in this file.

[SpexGuy]

Looks like the additional comptime execution for these tests might be putting the test build over the memory limit. That might be a reason to hold off on this until stage 2.

[andrewrk]

Landed in 5cc2e50 but with the comptime execution commented out in the tests. Idea being that we will not have test coverage of this for now, and re-enable them once we ship stage2.