Try to improve numerical precision of dynbal water and energy adjustments

[billsacks]

This is motivated by seeing some relatively large changes in the dynbal water and energy adjustments (QFLX_LIQ_DYNBAL and EFLX_DYNBAL) that I'm pretty sure are arising just from changing wa_col from 5000 to 4000 (a possibly-unintentional change in #523 ).

Currently, the dynbal water and energy adjustments operate by summing total water and energy before and after the changes in area. However, this means that changes in relatively small pools have their numerical precision compromised by potentially-unchanging, large pools. One such large pool is wa_col, but this problem probably occurs to a lesser extent when we have changes in canopy water with no changes in the much larger soil water states.

One way we could deal with this problem is to take differences of each state variable separately, then add those differences. Thus, for example, if wa_col isn't changing, then its difference would be 0, rather than swamping the differences in the terms that are changing.

If we do more rigorous water and energy conservation with dynamic landunits, as suggested by #274 , then this may become irrelevant, or at least less important.

[billsacks]

On further reflection, this could be a pain: If we're going to take differences of each state variable separately, we might as well just go with #274 . We could do a compromise, like breaking the total water down into a few categories based on the general magnitude of various state variables. But, given all sorts of other priorities, I can't see that fixing this is necessarily high priority right now, so I'm going to close it as wontfix.