Remove duplicate calculations of ice impedance

_From @billsacks on September 15, 2017 12:38_

@swensosc pointed out (a couple of months ago) that it would be good to remove duplicate calculations of ice impedance. These are things like:

```fortran
10._r8**(-e_ice*top_icefrac)
```

```fortran
10._r8**(-e_ice*(icefrac(c,1:3)))
```

and other variations on this. I think we can find all of these by searching for `e_ice`.

I did a quick search, and – while I think this is a worthwhile effort – I think it's going to take some effort. This is because there are many variations on this theme, and so some thought will be needed regarding how, exactly, to do this. I feel like it will take a full day or more to do this. So, given that this doesn't seem like the highest priority right now, I plan to defer this, unless others feel it should be done sooner.

Possible solutions I can see are the following - listed roughly in order from most efficient (but possibly hardest to accomplish) to least efficient (but possibly easiest to accomplish):

1. Calculate ice impedance just once each time through the run loop, and reuse that already-calculated value everywhere it's needed. However, I suspect that a single stored ice impedance variable isn't going to be sufficient, because of the many variations on the ice impedance calculation. So we may have to calculate a few versions of ice impedance to handle these different variations (also possibly using one of the other solutions below to remove duplication). Also, some analysis and/or testing is needed to see if it will work to just compute ice impedance once, near the start of the time loop, or if some code expects an updated ice impedance based on an updated icefrac (i.e., updated in the middle of the time loop).

2. Have an ice impedance routine that operates on arrays (within a filter). Then this could be called from the places that need it. This will require refactoring code that computes ice impedance, pulling the ice impedance calculation outside the loop.

3. Have an ice impedance routine that operates on single points. There already is such a routine in SoilWaterMovementMod, so this would need to be made public (either here or somewhere else). This will kill vectorizability if it's called from other modules, so only should be done as a last resort. However, it's possible that this could be combined with (1) without too much loss of efficiency - particularly if (1) involved just computing a few different versions of ice impedance all in a single module (in which case the single-point IceImpedance routine could potentially be inlined).

_Copied from original issue: NCAR/clm-ctsm#18_