Issue with 2 moment cloud microphysics, code shows grid or fails layout regression (but does not show grid pattern)

Just making an issue for this problem since it was not done (sorry I the title is not right, but I'm not an atmospheric scientist, all these difference "moist" schemes are just gobblygook to me). This is not a new issue, user @atrayano has been looking at it with fine tooth comb I'm told.
Two users @dbarahon and @wmputman have been see "odd" patters in some of the moist exports when running the 2 moment moist physics (I assume it is two moment because they have this CLDMICR_OPTION: MGB2_2M in the AGCM.rc file.
Both of them are seeing things like this it seems when examining output, this is "IWP" for example (this picture was made with ncview, using the "low" option for the scale, @dbarahon said this is the easiest way to spot this):

Where the domain decomposition is clearly appearing in the field and just not physical (I don't know how else to describe this).
In order to reproduce here is what you have to do.
First checkout v11.1.1 of the model. Then update the GEOSgcm_GridComp repo to the feature/donifan/KrokMG3 branch. When I did the test the SHA code for GEOSgcm_GridComp I used was 41d6f34f035193794b454c807e8a6a61bc7f9610.
Once built clone experiment on discover /gpfsm/dnb78s1/bmauer/donifan_weirdness. The AGCM.rc has this for the most options which I assume is critical:

###########################################################
# BACM_1M microphysics options
# ----------------------------------------
#CLDMICR_OPTION: BACM_1M
###########################################################

###########################################################
# GFDL_1M microphysics options
# ----------------------------------------
#CLDMICR_OPTION: GFDL_1M
#HYDROSTATIC: .TRUE.
###########################################################

###########################################################
# MGB2_2M microphysics options
# ----------------------------------------
CLDMICR_OPTION: MGB2_2M
#MGVERSION: 3
ENTR_DP: 2.e-4
USE_FCT: 0.
###########################################################

###########################################################
# shallow cumulus options
# ----------------------------------------
SHALLOW_OPTION: UW
###########################################################

###########################################################
# convection scheme options
# ----------------------------------------
CONVPAR_OPTION: GF
USE_GF2020: 1
# Convective plumes to be activated (1 true, 0 false)
DEEP: 1
SHALLOW: 0
CONGESTUS: 1
# Choice for the closures:
# deep : 0 ensemble (all) , 1 GR, 4 ll omega, 7 moist conv, 10 PB
# shallow : 0 ensemble (Wstar/BLQE) , 1 Wstar, 4 heat-engine or 7 BLQE
# congestus: 0 ensemble (Wstar/BLQE/PB), 1 Wstar, 2 BLQE, 3 PB, 4 PB_BL
CLOSURE_DEEP: 0
CLOSURE_SHALLOW: 7
CLOSURE_CONGESTUS: 3

Note I'm also told that @wmputman has feature branch that works off of develop branches off the GEOSgmc fixture: feature/wmputman/hwt_spring_exp but I don't want to record anything here for reproducibility that depends on a branch since branches change...

I thought user @dbarahon claimed that downgrading the optimization to O1 in the moist component fixes (maybe I misunderstood) this but that's not what my testing showed so if there was a workaround that was not it. Whether I compiled those files in moist O3 or O1 I got these weird patterns clearly showing the domain.