Reset configuration of 3DFGAT atmos

[rtodling]

Presently the 3D-FGAT flavor implemented in SWELL is triggered in the following configuration:

1. it uses the 4D-Var cost type
2. it uses hourly backgrounds

I would like to changes this to:

A) use 3D-FGAT cost type
B) use 3-hourly backgrounds

From a JEDI perspective these changes are trivial - I have the yaml settings needed to get the var to run in this context; but we'll need a little doing in SWELL, particularly when it comes to handling the increment to EVA. The issues is that using the 4D-Var cost type to handle 3D-FGAT results in an increment being put out in the beginning of the time window, for example, for a 00z cycle, the increment is written 21z of the previous day; whereas in the context of 3D-FGAT cost type JEDI write the increment out in the middle of the var window (ie, 00z in the example). I think the work-flow needs a little adjustment to adapt to this change.

For consistent configurations of the Var the changes above result in the following differences in timing and memory usage:

Current: OOPS_STATS Run end - Runtime: 2883.52 sec, Memory: total: 1413.65 Gb
Revised: OOPS_STATS Run end - Runtime: 1299.55 sec, Memory: total: 772.22 Gb,

Where it should be noted that neither of these use exactly the default configuration in SWELL for atoms-var. The configurations above use c90 background,. c180 background error covariance, 216 PEs.

The changes proposed here make the Var configuration in JEDI more consistent w/ GSI's, and it should be pointed out that the changes in results from the present configuration to the one proposed here are small in comparison to the differences we have been observing between the two systems.