Bring back dual resolution covariance capability

[rtodling]

When I removed the ESMF interface to handling the ensemble I handled things in the most obvious ways - using the underlying distribution arrays available for the background grid. That makes things easier when reading the ensemble, but it prevents the climatological and ensemble background errors to have different resolutions.

Use of single resolution would require either the ensemble to have to increase in resolution to so we don't change the clim-B resolution or vice-versa, a reduction in resolution of clim-B - either of these would make it hard to get a clean comparison w/ GSI's results.

This issue is a reminder that this must be reinstated.