Add a test that ensures that setting all_active = .true. doesn't change answers for gridcell averages

**Bill Sacks < sacks@ucar.edu > - 2016-06-13 20:37:51 -0600**
**Bugzilla Id:** [2322](http://bugz.cgd.ucar.edu/show_bug.cgi?id=2322)
**Bugzilla CC:** erik@ucar.edu, rfisher@ucar.edu,

Changing which 0-weight points are active should not change answers for gridcell averages. If this does change answers, it indicates a bug: values in 0-weight patches or columns should not affect the gridcell averages.

I have found two bugs by doing a manual test like this: bug 1851 and bug 2321. (See notes in bug 1851 for more detailed thoughts.)

We should add an automated test of this. Specifically, it could do two runs: one out-of-the-box and one with all_active = .true. (probably with finidat = ' ', because it can be problematic to set all_active = .true. with an finidat file not set up for that situation - although maybe we could use an initial conditions file along with init_interp?) The gridcell-level history fields should be bit-for-bit identical in these two runs. We could do that with a mechanism similar to the current LII test. However, this will hopefully become more straightforward once https://github.com/CESM-Development/cime/issues/146 is resolved - hopefully making it relatively trivial to create one-off tests like this.