Avoid doing dyn_subgrid calculations if there were no changes in subgrid areas and/or only allow dynamic landcover on year boundaries

In timing studies, `dyn_subgrid` takes 1-2% of the total runtime even if there are no changes in area. I could add a check for whether there were any changes in area in this time step, and if not, avoid doing anything else.

Update (2020-07-17): Another thing we could do is to bake in the assumption that changes only happen on the year boundary, unless some flag(s) are set saying that mid-year changes are possible. Then (other than the dribbler stuff noted below) I think we could avoid doing any work for the vast majority of time steps. (Update (2020-08-17) this would be important if we do the memory overhaul in https://github.com/ESCOMP/CTSM/projects/32.)