Fix calculation of COA times for bistatic images

[Klamptoo]

SARkit was returning different scene_to_image() results depending on the contents of the input scene_points array. If it contained one point, the returned image location was correct. If it contained a list of scene points, the returned image locations were slightly wrong. I tracked it back to the norm calculation in compute_coa_pos_vel().

With the change, this simple test now passes:

scene_coords = np.random.default_rng(789).uniform(low=-240.0, high=240.0, size=(7, 3)) + scp
assert(np.all(sksicd.scene_to_image(sicd_xmltree, scene_coords[0:1,:])[0][0] == sksicd.scene_to_image(sicd_xmltree, scene_coords)[0][0]))

[pressler-vsc]

Nice find; thanks for taking the time to track this down and open a PR