Update MPAS-O: bug fixes for KPP nonlocal source term

[mark-petersen]

This PR fixes several bugs in the calculation of the nonlocal source term in KPP. for temperature, we need to subtract off the contribution from thickness fluxes that don't change buoyancy. for salinity, we need to add virtual salinity fluxes that do change buoyancy.

These fixes have been tested in E3SM G-cases and behave as expected.

Fixes #3102
[non-BFB]

[mark-petersen]

@maltrud, this is a placeholder for the moment. When MPAS-Dev/MPAS-Model#305 is merged, this will show the correct diffs.

[rljacob]

Please point to the github issue reporting these bugs.

[maltrud]

@rljacob i don't think there is an issue--just something @vanroekel and i have been sleuthing out.

[rljacob]

Ok. Usually when someone says a PR "fixes several bugs" we'd like at least one issue summarizing the problem or how it manifested.

[mark-petersen]

PR now points to bug issue. @jonbob, this is ready to merge onto next.

[jonbob]

@rljacob - are you satisfied with the "bug fix" part of this now?

[rljacob]

yes

[jonbob]

merged to next

[jonbob]

merged to master and expected DIFF's have been blessed -- except for:

• SMS_Ld2.ne30_oECv3_ICG.A_WCYCL1850S_CMIP6.anvil_intel.allactive-v1cmip6
  which is failing because of a testing problem downloading the cvmix code (but otherwise
  would have an expected DIFF);
• SMS_Ld2.ne30_oECv3_ICG.A_WCYCL1850S_CMIP6.cori-knl_intel.allactive-v1cmip6
  which is failing due to machine issues (but otherwise would have an expected DIFF)

[jonbob]

blessed the DIFF for SMS_Ld2.ne30_oECv3_ICG.A_WCYCL1850S_CMIP6.anvil_intel.allactive-v1cmip6 now that the testing system is downloading the cvmix code again