Measuring centrality and dispersion in directional datasets: the ellipsoidal cone covering approach