A talk with the experts on the OGC [Coordtran.wg] mailing list teach us that we should really not allow ellipsoidal height to exist independently of the (latitude, longitude) coordinates, and in the case of a projected CRS it should not be allowed at all. Consequently we need to delete VerticalDatumType.ELLIPSOIDAL, which was imported from the legacy OGC 01-009 specification.
Removing that particular vertical datum type may be an issue for WKT parser, since this datum type do exists in the WKT specification. However we can leave the definition of whatever placeholder for "ellipsoidal height" on the implementor side. The issue here is to remove it from public API, but it may stil exists internally in some implementation.
On the API side, this cause an issue with VerticalExtent.getVerticalCRS() in metadata, which will be unable to return an extent relative to the ellipsoid unless we revert the return type from VerticalCRS to a generic CoordinateReferenceSystem as it was in the original ISO 19115 specification. But according the mail exchange, the intend is really to prevent such API to use ellipsoidal height.