optimized island cleanup to 0.25x the cost, flat 2 pointer array instead of a NativeQueue and replacing frequent coord->index conversion for index stride offsets