NaN in nitrogen_dynamics when running with land cover change

[luyssaert]

When running r6265 on a simple 3 by 3 test case, the model crashes on an NaN problem in LCC.

0 Date: 1986-01-02 0.0000sec at timestep 1489249
0 At the pixel 2 for element 2 and for PFT 2
0 pool_end is NaN
0
0FATAL ERROR FROM ROUTINE nitrogen_dynamics
0 --> NaN problem

[luyssaert]

The PFT has been removed but there is still some water on the canopy a solution need to be found for this water. If it is a forest PFT that was removed we will just add the water to soil water column of the tall vegetation. Note that it is also possible that last forest was removed. In that case there is no longer a tall vegetation water column. In that case we need to find a different water column to add the canopy water to. Ideally that would be to water column to which the new PFT belongs. For example if the last forest became a cropland the water previously stored in the forest canopy should be added to the soil water column of the short vegetation. Because the current land cover change functionality only deals with net land cover changes we don't know the exact changes. An approximation (add to the largest soil column) was added in r6273.