Add INIT_BOUNDARY_EXTRAP parameter and function to allow ALE boundary extrapolation behaviour to differ at initialisation and in model run

[claireyung]

Adds INIT_BOUNDARY_EXTRAP parameter and function, so that REMAP_BOUNDARY_EXTRAP can just be used in the initialisation and not in the model dynamics, as required for an ice shelf in ALE mode.

Background:
When an ice shelf is initialised in ALE mode, MOM_state_initialization first initialises the ocean thickness and T/S with topography, but without an ice shelf. It then calls the function trim_for_ice, which takes in input of the ice shelf pressure, and in turn calls cut_off_column_top which truncates the columns (and adds vanishing layers) to account for the pressure of the ice depressing the free surface. The output of this is a grid with the ice shelf boundary treated in a ALE z-coordinate fashion with vanishing layers at the surface, and @adcroft's upcoming changes will make T and S consistent to the grid. However, after adding the ice shelf, we still need to regrid and remap onto our desired ALE coordinate, for example sigma coordinates. This can be done with regrid_accelerate in MOM_state_initialization or by the block controlled by REMAP_AFTER_INITIALIZATION in MOM.F90

We want the initialisation to use boundary extrapolation in the remapping so that the boundary cells preserve the desired initialisation profile when remapped to their target coordinate. This is controlled by REMAP_BOUNDARY_EXTRAP, except this runtime parameter modifies the control structure for the dynamic part of the model run too, which is bad because it means extrema are not preserved in the ALE remapping and can lead to non-sensible values of tracers.

Changes:

• Added runtime parameter INIT_BOUNDARY_EXTRAP and set the ALE control structure to use that instead of REMAP_BOUNDARY_EXTRAP at first. The default of INIT_BOUNDARY_EXTRAP is REMAP_BOUNDARY_EXTRAP to preserve answers (though it is not recommended to use REMAP_BOUNDARY_EXTRAP during model dynamics...)
• After ALE regrid/remap in MOM.F90 (final initialisation step using ALE), reset the boundary extrapolation flag in the ALE control structure to go back to REMAP_BOUNDARY_EXTRAP.

This PR combined with @adcroft's upcoming changes should make the initialisation of the ocean thickness and TS in an ALE ice shelf cavity perfect, when the new flag INIT_BOUNDARY_EXTRAP = True and old flag REMAP_BOUNDARY_EXTRAP = False.

[Hallberg-NOAA]

This is a little bit complicated, but it is explained very well in the description, and it should lead to the intended improvements in capabilities.

[marshallward]

Gaea regression: https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/23284 ✔️ 🟡

Hash was slightly out of date but going ahead with this anyway.

Squash-merge this one.