Fix the both-null guard in SorterBase to test the right-hand side - #9098

In SorterBase<T>::compare_vfunc the "both sides failed to cast" guard tests concrete_lhs twice, so it never looks at the right side at all. The two g_return_val_if_fail lines under it are written to separate the one-sided cases, and the first of them can't be reached the way it stands: with concrete_lhs null and concrete_rhs fine, the guard already fired and returned EQUAL instead of SMALLER.

This only shows up on the GTK4 path when Glib::wrap_auto hands back something that isn't a T, which is why nothing has complained about it. The names sit next to each other on consecutive lines, which is probably how it happened.