Fix issue 1357: Zeros in unscented covariance

This fix is for issue #1357 (closed) and simply changes getEvolution in UnscentedKalmanModel to copy across the non-orbit parts of the input sigma points into the predicted sigma points.

I needed to change the initial covariance matrix in UnscentedKalmanModelTest because it wasn't positive definite. I've just changed it to be diagonal.

I haven't added a new test for this bug. Would that be a good idea?

edit: This has turned into a much more significant change. This merge request will enable the UKF to estimate both propagation and measurements parameters.