landuse_transition_matrix is uninitialized for non-land use run modes

[glemieux]

Original issue discovered in ESCOMP/CTSM#2656, which was overlooked due to ESCOMP/CTSM#2657. I think we just need to initialize this to zero in EDInitMod via zero_site (as we do with site_in%disturbance_rates).

Note that this appears while using the gnu compiler on derecho. Other compilers on derecho initialize to zero.

• check if this shows up with gnu on derecho without debug
• check if this shows up on perlmutter with gnu

[glemieux]

Noting this here for posterity: the garbage output is also showing up with land use mode on for the first time step (written during initialization, but prior to fates dynamics_driv call). That said, the matrix is appropriately updated on the second step (i.e. first dynamics update). So this technically affects all run modes, not just satellite phenology, but only during the first time step. Since we avoid disturbance_rates in sp mode, this shows up as unitialized for every time step.