Modularize thermodynamics

[benegee]

We have many different compressible Euler equation systems in Trixi.jl and TrixiAtmo.jl by now. It would be good to have a common structure and modularize the thermodynamic components such that they could be easily swapped out.

[MarcoArtiano]

That might also concern the initial conditions. If they are given in terms of the primitive variables, then we do not need to specify and repeat the initial conditions for each set of equations. For example, right now the baroclinic instability has duplicate functions for CompressibleEulerPotentialTemperatureEquationsWithGravity3D and CompressibleEulerEnergyEquationsWithGravity3D.

[benegee]

Absolutely right!
@tristanmontoya already made some suggestions:

• implement reference states / initial conditions in a central place
• settle on a set of variables, maybe primitive
• in elixirs call transform_initial_condition to transform to the set of variables to the ones actually used
  (this additional step might just be a call to prim2cons, but is required for the covariant solver, where there is more work to do)