ShonanAveraging Covariance fixes

[johnwlambert]

• use upper 3x3 sub-block of covariance matrix for converting BetweenFactor to BinaryMeasurement, instead of erroneous lower 3x3 sub-block, as the tangent space of SE(3) is ordered as (R,t)
• Use Isotropic in ShonanAveraging2, instead of taking a 1x1 subblock of a matrix

In GTSFM, we use the erroneous 3x3 lower-right sub-block, but it doesn't matter currently because we set each dimension to a unit Gaussian, as follows:

        noise_model = gtsam.noiseModel.Unit.Create(6)

        between_factors = gtsam.BetweenFactorPose3s()

        for (i1, i2), i2Ri1 in i2Ri1_dict.items():
            if i2Ri1 is not None:
                # ignore translation during rotation averaging
                i2Ti1 = Pose3(i2Ri1, np.zeros(3))
                between_factors.append(BetweenFactorPose3(i2, i1, i2Ti1, noise_model))

        obj = ShonanAveraging3(between_factors, shonan_params)