ifx KGO changes from v006 to v007 are unphysical

[jshaw35]

Changes to the KGOs to account for ifx compiler updates contain unphysical changes that appear to mask model/compiler errors. Those changes are summarized in the PR here. Many differences appear to be at the roundoff level, but some variables have larger changes that are unphysical rather than resulting from changes in masking values. Specifically, the affected variables are in the ISCCP simulator (cltisccp, pctisccp, tauisccp, albisccp) and the MISR cloud histogram field (clMISR).

Plotting the ISCCP total cloud fraction alongside the similar MODIS field for v006 and v007 shows these non-physical changes. The highest cloud fraction values are reduced, and the artificially imposed maximum value changes between processing chunks (index 1000).

I cannot reproduce this behavior locally, but adding print states to diagnose it in the CI produces similar changes in the outputs. See ifxerrors. This could explain why additional code from COSP-RTTOV creates similar differences.

These errors appear to depend on both: 1. The volume of changes to the code base (adding one additional print statement preserved tests but adding ~10 did not), 2. The size of the data (errors appear to increase between the basic and global tests).