Patches to BT-solver for ice-shelf layer configs

[adcroft]

The ice-shelf configurations push the BT solver to new extremes and have been cluttered by "eta has dropped below bathyT" WARNINGs. This turns out to be because the linearization of BT transports about the baroclinic thicknesses is far from accurate in the vicinity of the grounding line. Rather than address the linearization (which is hard to improve on) we apply a flux limiter that stops divergent fluxes from removing mass that does not exist. Because the reconstructions are fixed, this is done as a flux limiter rather than adjustments to the reconstruction (as is done in the layer continuity fluxes). This BT flux limiter has to account for sinks that will be applied over future BT steps, which means that there are situations where flooding from one direction does not allow draining in another, e.g. where the sink uses up the entire column. The rate at which sinks was applied in the BT solver was leading to negative thicknesses in some cases. We also found that recalculation of the sinks (corrections) was detrimental and made the negative-thickness more likely.

There are three main fixes:

• Change the rate at which sources/sinks are applied throughout the BT solver, to avoid grounding out during the filter steps. Fix enabled with BT_ADJUST_SRC_FOR_FILTER = True
• Disable the re-calculation of BT mass source after the predictor step, using BT_ADJ_CORR_MASS_SRC = False
• Apply a flux limiter based on initial volume adjusted by mass sinks, enabled with BT_LIMIT_INTEGRAL_TRANSPORT = True

With the three fixes enabled, a grounded ice-shelf test with active sources/sinks was able to initialize and run without negative thicknesses occurring for 15+ days.

Incidental changes:

• One minor repositioning of an hchksum() encountered when debugging layout problems (which still exist).
• Added checks on initial thicknesses that are passed to the BT solver
• Made negative thickness checks FATAL instead of just WARNINGs, when using the limiter. There is no circumstance I can think of where such a non-physical state should be allowed.

Notes:

• I tried making the non-negative check on eta_IC always active and FATAL but incredibly "double_gyre" fails! In the interests of progress, I opted to make the test conditional on the limiter being active.

[Hallberg-NOAA]

I agree that these changes are correct.

I would note, however, that the description of this PR is wrong on one point. The transports in the barotropic solver are not a "linearization" of the layer transports. Rather they are a nonlinear parametric fit to the transports that only amount to a linearization in the limit of small barotropic velocity changes.

[Hallberg-NOAA]

Resolving adcroft#6 will automatically address the one omission in this PR that is needed before it can be accepted.

[Hallberg-NOAA]

This PR is now ready to be merged into dev/gfdl.

[Hallberg-NOAA]

This PR has passed pipeline testing at https://gitlab.gfdl.noaa.gov/ogrp/mom6ci/MOM6/-/pipelines/28465 with the expected warnings about new runtime parameters.