Potential issue in equivalent buoyancy formulation of the ThermalSWSolver class

[nhartney]

This flags a possible issue in the formulation of the ThermalSWSolver when using the equivalent buoyancy formulation. The notation uses the variable b to represent both buoyancy and equivalent buoyancy. In either formulation (equivalent buoyancy or not) the field actually used in the solver should always be b. Then when using the equivalent buoyancy formulation the b_e field recovered from the equation set probably should be converted to b before being used by the solver. This currently happens with the reference profiles but not the fields.

[nhartney]

Other things that might have changed, so that we have a record:

• The original MoistDynamicsSWSolver used the input fields xn rather than the reference fields X_ref in calculating the saturation function and the water vapour - we think that the current approach of using X_ref should be better.
• Previously the reference profiles were not updated at each timestep and currently they are.

[tommbendall]

I had been hoping this would be resolved by generating all of the linearisations automatically in #649, but I haven't been able to remove all of the hand-coded linearisations.

I don't suggest we make this change though without first doing a thorough set of tests with the equivalent buoyancy formulation of the moist thermal shallow water equations. So I think we should leave this PR here until we've had time to do that

[jshipton]

phew!