Major refactor of FieldBoundaryConditions

[glwagner]

This PR introduces a major refactor of FieldBoundaryConditions and changes the user API.

The grid and field location are no longer necessary to build FieldBoundaryConditions. This means that the constructors TracerBoundaryConditions, UVelocityBoundaryConditions, etc, are gone. Instead we have only FieldBoundaryConditions, which is used like

u_bcs = FieldBoundaryConditions(top=FluxBoundaryCondition(-1e-4), bottom=FluxBoundaryCondition((x, y, t, u) = -1e-3 * u, field_dependencies=:u))

T_bcs = FieldBoundaryConditions(top=FluxBoundaryCondition(1e-8))

There's also a new constructor AuxiliaryFieldBoundaryConditions which takes the place of ComputedFieldBoundaryConditions and is now the default for Field and ComputedField. The difference is that AuxiliaryFieldBoundaryConditions have nothing instead of an ImpenetrableBoundaryCondition in Bounded directions when they're located on faces.

Under the hood the FieldBoundaryConditions struct is "flattened" so there's no more CoordinateBoundaryConditions. In addition, a new field immersed is added to FieldBoundaryConditions in addition to east, west, etc. This will be used to support boundary conditions on immersed boundaries in the future.

There's still quite a few validation experiments to clean up and probably some lingering issues with tests, but I was able to get the examples to run so I thought it was the right time to open the PR.

On the API changes: it seemed like a positive change that we don't need grid in the boundary conditions constructor, and that we only have one name FieldBoundaryConditions rather than 4. However, if others feel they liked the old API, we can add convenience functions TracerBoundaryConditions, etc back to the source.

[navidcy]

Ok, this definitely needs bumping up a minor release

[navidcy]

Great job. I’ll try to help out cleaning up the rest.

[navidcy]

@glwagner, regarding the

ERROR: LoadError: UndefVarError: bc not defined

in the Docs, is it related to
JuliaDocs/Documenter.jl#228
perhaps?

[navidcy]

I believe it should be OK now. Commit e39c509 was the crucial fix here.

[navidcy]

I approve.

I applaud the tests/jldoctests.

I'm a bit skeptical whether I have the expertise to be as critical.

[navidcy]

https://clima.github.io/OceananigansDocumentation/previews/PR1843/

[glwagner]

@glwagner, regarding the

ERROR: LoadError: UndefVarError: bc not defined

in the Docs, is it related to
JuliaDocs/Documenter.jl#228
perhaps?

Nice find!

[glwagner]

I'm going to clean up the validation experiments for the new syntax and then this is ready.

[glwagner]

I approve.

I applaud the tests/jldoctests.

I'm a bit skeptical whether I have the expertise to be as critical.

The main point of approval here is whether we think the API change is positive. I think you are one of the best people anywhere to judge this change.

The other change, which is adding a field immersed, is innocuous. We have to have some way of implementing boundary conditions on immersed boundaries and this is one solution to that. Probably its hard to say for anyone what the "best" solution is (but alternative designs are welcome).

Note that it's possible to implement a field immersed without the API change. But I felt the API change was positive / important (I feel we've got to squash technical debt ASAP in all cases or I think it will bite us worse in the future).

[navidcy]

We should wait for this to be merged before we move on with #1847, right?

[glwagner]

When we merge this we will have to update the example in #1847 but otherwise this is just an aesthetics change and has no impact on output, so we don't need to wait for it.