Support non-monotonic diagnostic grids

[ashao]

@StephenGriffies, @gmacgilchrist and I have been discussing support for diagnostic grids which are non-monotonic in the model's native space. The continuous isopycnal coordinate is one such case if density is non-monotonic, but others (such as potential/conservative temperature) also have useful applications. I will note that currently, the diagnostic continuous isopycnal coordinate does not have well-defined behavior in the presence of density inversions because 'convective adjustment' is not done (though the prognostic version of the model does).

In order to derive a well-defined grid and close budgets for extensive variables the following must be done:

1. At the beginning of the timestep, each potentially non-monotonic coordinate must calculate and save the re-ordering of the native k-space that makes diagnostic coordinate monotonic.
2. The target grid is then constructed based on this reordering
3. When a diagnostic is posted, the column must be rearranged according to the saved re-ordering and 'reintegrated' onto the target grid
   A similar methodology needs to be done for intensive quantities.

This has the potential for increasing the cost of diagnostics for the model because sorting can be very inefficient and an additional copy of the model's native column must be retained.

[StephenGriffies]

Even though potentially quite costly, this diagnostic would have heaps of very cutting-edge analysis applications and put MOM6 well ahead of other models in its capabilities for analysts. Let's go for it!