wrong output for genetic_relatedness(... indexes=None, proportion=True)

[petrelharp]

These should be the same, except with one less dimension. They are not.

> ts.genetic_relatedness(sample_sets=[[0, 1, 2], [3]], indexes=[(0, 1)], windows=[0, 5, 10])
array([[-0.11970621],
       [-0.122826  ]])
> ts.genetic_relatedness(sample_sets=[[0, 1, 2], [3]], indexes=None, windows=[0, 5, 10])
array([[-0.11970621, -0.13156656],
       [-0.11175358, -0.122826  ]])

The issue here was our naive normalization code was dividing (in this case) a shape (2, 1) array by a shape (2,) array, which is a shape (2,2) array, not (2,1) as desired.