Fixed a recently added bug in insert_brine

  Fixed a recently bug in insert_brine that was introduced yesteray with MOM6
commit 965c2c2.  This bug is a double inclusion of US%s_to_T, so technically
it does not change answers, just breaks dimensional consistency testing.  For
some reason the impacted code was not triggered by the MOM6-examples test cases.

src/parameterizations/vertical/MOM_diabatic_aux.F90

@@ -429,7 +429,7 @@ subroutine insert_brine(h, tv, G, GV, US, fluxes, nkmb, CS, dt, id_brine_lay)
     salt(:)=0.0 ; dzbr(:)=0.0
 
     do i=is,ie ; if (G%mask2dT(i,j) > 0.) then
-      salt(i) = US%s_to_T*dt * (1000. * US%R_to_kg_m3*US%Z_to_m*US%s_to_T*fluxes%salt_flux(i,j))
+      salt(i) = US%s_to_T*dt * (1000. * US%R_to_kg_m3*US%Z_to_m*fluxes%salt_flux(i,j))
     endif ; enddo
 
     do k=1,nz