Add methods to average quaternions

[janbridley]

Methods for averaging quaternions were added, based on Markley 2007, who provides a maximum likelihood method for the average quaternion. This is different from the standard iterative (Weiszfeld) method for averaging rotations, and is faster and simpler (but not necessarily more correct?)

Quick and dirty benchmarks:

# let q_N = rowan.random.rand(N)
>>> %timeit -r 100 -n 1000 mean(q_1)
5.73 μs ± 1.41 μs per loop (mean ± std. dev. of 100 runs, 1,000 loops each)

>>> %timeit -r 100 -n 1000 mean(q_10)
6.01 μs ± 1.38 μs per loop (mean ± std. dev. of 100 runs, 1,000 loops each)

>>> %timeit -r 100 -n 1000 mean(q_100)
6.17 μs ± 1.45 μs per loop (mean ± std. dev. of 100 runs, 1,000 loops each)

>>> %timeit -r 100 -n 1000 mean(q_1000)
8.38 μs ± 1.81 μs per loop (mean ± std. dev. of 100 runs, 1,000 loops each)

>>> %timeit -r 100 -n 1000 mean(q_10_000)
29.8 μs ± 2.71 μs per loop (mean ± std. dev. of 100 runs, 1,000 loops each)

>>> %timeit -r 100 -n 1000 mean(q_100_000)
244 μs ± 4.31 μs per loop (mean ± std. dev. of 100 runs, 1,000 loops each)

TODO:

• Properly cite paper
• Add iterative refinement? Verify how much this changes answers