FCT is not monotonic with non-linear free surface

Running the LOCK_EXCHANGE test case, I noticed that minimum and maximum temperatures do not stay within the bounds set by the initial state (e.g. between 5 and 30°C). This is illustrated in the plot below with small, but abnormal deviations, considering that FCT should avoid this. This only occurs with non linear free surface, so based on this, and after digging a bit in the nonosc routine, I reached the conclusion that the scale factor used in the flux partitioning is not at the right time level here. It should be taken at "after" time step (i.e. Khrs as available in the code, or for better readability Kaa).