Bug fix (bug #44)

Doug Jacobsen points out that MPAS-O differentiates between total number of
vertical levels and active number of vertical levels... so the array
CVmix_vars%Mdiff_iface may have more than CVmix_vars%nlev+1 elements. The
subroutine cvmix_coeffs_kpp_low did not consider the situation where Mdiff_out
was a smaller array than old_Mdiff.

src/shared/cvmix_kpp.F90

@@ -596,10 +596,13 @@ subroutine cvmix_coeffs_kpp_low(Mdiff_out, Tdiff_out, Sdiff_out, zw, zt,    &
     interp_type2   = CVmix_kpp_params_in%interp_type2
     MatchTechnique = CVmix_kpp_params_in%MatchTechnique
 
+    nlev_p1 = size(Mdiff_out)
+    nlev    = nlev_p1 - 1
+
     ! Output values should be set to input values
-    Mdiff_out = old_Mdiff
-    Tdiff_out = old_Tdiff
-    Sdiff_out = old_Sdiff
+    Mdiff_out = old_Mdiff(1:nlev_p1)
+    Tdiff_out = old_Tdiff(1:nlev_p1)
+    Sdiff_out = old_Sdiff(1:nlev_p1)
 
     ! (1) Column-specific parameters
     !     
@@ -609,9 +612,6 @@ subroutine cvmix_coeffs_kpp_low(Mdiff_out, Tdiff_out, Sdiff_out, zw, zt,    &
     kwup = floor(kOBL_depth)
     ktup = nint(kOBL_depth)-1
 
-    nlev_p1 = size(zw)
-    nlev    = size(zt)
-
     if (ktup.eq.nlev) then
       ! OBL_depth between bottom cell center and ocean bottom, assume
       ! zt(ktup+1) = ocn_bottom (which is zw(nlev+1)