BkgErrorGODAS needs a QC step

[guillaumevernieres]

Description

This is the parametric background error that we have used for ... ever ... And while it "does the job" for the most part, it needs a QC step or a bit of TLC so we don't end up with fields like this

To Reproduce:

I don't know if this would happen at 1deg, probably.
With the 1/4 deg MOM6, bump up the amplitude of the the diagonal of B, run one cycle and look at the background error and the analysis and you should see something like the figure linked above.
I'm pretty sure this is happening in the regional case as well (pinging @kbhargava).

Possible temporary solutions

We compute the diag of B separately in the EMC implementation (GDASApp) and added a horizontal filter to smooth the background error. I'm not convinced that smoothing garbage is the way to go though ...