h/mercury: Dynamically allocate arrays for stereographic projection
needs testing! I thought I saw this causing issues on my machine, but now I suspect it's the hand orientation guesser that was the real problem. If this works fine on a few people's machines I guess I'll shrug and merge it.