uniform quaternion generation computationally uneven across different numbers of orientations

[cmahrens]

We noticed a computational anomaly in skopi.get_uniform_quat(N_orientations, True). It spends far less time to do this for 64k or even 256k orientations compared to 8k, 16k, 32k and 128k. Any idea why? Thanks.

>>> def doquat(num):
...     start = time.time()
...     skp.get_uniform_quat(num,True)
...     end = time.time()
...     elapsed = end - start
...     return elapsed
... 
>>> doquat(1024)
0.013123035430908203
>>> doquat(2048)
0.030680179595947266
>>> doquat(1024*8)
22.523311614990234
>>> doquat(1024*16)
44.15811800956726
>>> doquat(1024*32)
86.49000382423401
>>> doquat(1024*64)
0.8508529663085938
>>> doquat(1024*128)
337.60426592826843
>>> doquat(1024*256)
1.994347095489502