bug in order of operations for snow density

[ecoon]

Current master has a bug in the PVFE kept by surface_balance_implicit_subgrid PK. The PVFE for snow density is actually the OLD state, not the NEXT state. Therefore, when it is tagged as "changed," no change is actually registered.

This bug avoids a weird loop, whereby:

1. snow PK requests update of snow-source_sink
2. this triggers update of SEB evaluator
3. after this update, the snow PK changes snow-density
4. then the flow PKs request update of snow-source_sink, which, because it depends on snow density, is updated again.

It seems that this cycle should be broken somehow, but currently it is broken by a bug.