Add RK2 loop in time integration module. This is the bare bones of
RK2 time stepping. Compiles but likely will not run.

Pass rkTend to advection routine to accumulate tendencies

Correct dimensions of rkTend, which was inadvertently 2D instead
of 3D.

Update halos for rkTend in each step of RK loop

Ensure damage >= damageNye after RK integration. Previously,
li_finalize_damage_after_advection did not impose the damageNye minumum.

Generalize RK time integration to use order defined in namelist.
Currently supports config_rk_order = 1, 2, 4. Third order to be
added soon.

The previous Runge-Kutta commits left forward Euler time integration
non-functioning. This makes FE time integration function again.

Rename files, modules, and routines to include Runge-Kutta

Add support for config_rk_order = 3, using the Strong Stability-
Preserving RK3 method. This follows a slightly different algorithm
from the classic RK2 and RK4 methods, and so requires significantl
increasing the complexity of the RK loop. See eq 2.48 (pg 56) in
Durran (2010): Numerical Methods for Fluid Dynamics for the description
of this method.

Previous commits inadvertently updated sub-step solutions incrementally
(i.e., using y(k-1) to solve for y(k), where k is the substep), when they
should have been using y0. This commit fixes that, although tests show
that this treatment is not complete and solutions are unstable.

Only use SSP RK2 and RK3 methods, and remove support for classical
RK2 and RK4.

Rename some local variables, update comments, update log write

Add option for 4-stage SSPRK3, which involves four velocities solves
instead of three, but also allows for a CFL fraction of 2. To use
the 4-stage formulation, set 'config_rk_order = 3' and
'config_rk3_stages = 4'.

Include SMB and BMB within RK loop by calculating a weighted average
based on each specific RK formulation.

Fix SMB and BMB budget for forward euler, as previous commit caused
these fields to be written out as all zeroes.

Correct a description in Registry, update comments and log writes.

Fix bug in groundedSfcMassBalApplied after Runge-Kutta loop. It was
inadvertently being assigned the enitre applied SMB field instead
of just the grounded portion.

Calculate the total groundedToFloatingThickness and fluxAcrossGroundingLine
at the end of the RK loop. Budgets do not yet close, but are reasonably close.

Reset layerThickness when using config_restore_thickness_after_advection.
Restoring thickness alone is not sufficient when using Runge-Kutta
time integration because the advected layerThickness gets propagated through
all stages of the Runge-Kutta loop, which leads to advection and surface/basal
mass balances causing thickness change.

Update stresses and strain rates after calving. This allows
restart test with damage to pass validation, but may not be the
most rigorous fix.

Update thickness and tracer halos before velocity solve. This allows
eismint2 decomposition test to pass validation.

Update temperature and waterFrac in RK loop, but not enthalpy. Updating
enthalpy resulted in negative temperatures, which caused an error in li_thermal.
Updating temperature and waterFrac instead after enthalpy is advected allows
enthalpy tests in full_integration suite to pass execution and validation.

Simple clean-up after code review, including moving call to mask
calculation, initializing waterFracPrev to zeroes, adding return
statements after error messages. Also add a halo update on cellMask
and edgeMask before calculating groundedToFloatingThickness and
fluxAcrossGroundingLineOnCells.

Initialize the various RK weights (rkSSPweights, rkTendWeights,
rkSubstepWeights) to large negative values which are overwritten
for each case. This will make it obvious if a weight is being used
when it should not be.

Restore calving front within RK loop to prevent velocity solutions
on extended ice geometry.

Remove support for config_rk_order = 1 to avoid unnecessary redundancy
with config_time_integration = forward_euler while maintaining
backward compatibility.

Include calving in Runge-Kutta loop, rather than implementing it
afterwards.

Calculate layerThickness after calving in RK loop, which is necessary
when calculating weighted averages for RK integration.

This reverts commit 5e23b98.

This reverts commit 813ee05. After
testing and discussion, it is apparent that including calving within
the Runge-Kutta loop is not desirable because it changes the domain
between stages of the Runge-Kutta integration.

Calculate fluxAcrossGroundingLineOnCells in advection module for each
RK stage and then take weighted average after RK integration instead
of calcuating fluxAcrossGroundingLineOnCells on final state. This is
more consistent with how fluxAcrossGroundingLine (on edges) is treated.

…alving

Add logic to control whether velocity is solved before and/or after calving.
This allows both for backward compatibility and for the abilility to
use self-consistent velocty, stress, and strain-rate fields when calculating calving.

Make config_update_velocity_before_calving = .false. the default. This
is not necessarily the desired behavior for production runs, but it
is necessary for baseline comparison when using forward Euler time
integration.

…ving.

Always update velocity after calving if it was not updated before calving.

… positions

Move subglacial hydro and bed topography updates back to their original positions.
These were initially moved due to the complexity of interaction with
RK integration, but I think subsequent changes have made that move
unnecessary.