rft: clean up hyperdiffusion

[haakon-e]

This pull request refactors and streamlines the hyperdiffusion logic in src/prognostic_equations/hyperdiffusion.jl, focusing on code clarity, maintainability, and consistency.

Key improvements and changes:

• extract components from parent structs (e.g. moisture_model from atmos, or ᶜu from p.precomputed) to simplify and shorten their inline usage
• combined multi-line assignments and expressions, making code more concise.
• give new names to 4th order derivatives to clarify their usage.
  For example, instead of the (somewhat confusing) expression

@. ᶜ∇²u =
        divergence_damping_factor * C123(wgradₕ(divₕ(ᶜ∇²u))) -
        C123(wcurlₕ(C123(curlₕ(ᶜ∇²u))))
@. Yₜ.c.uₕ -= ν₄_vorticity * C12(ᶜ∇²u)

I propose assigning a new name to the resulting calculation. Importantly, this does not affect the underlying memory of the data

ᶜ∇⁴u = @. ᶜ∇²u =
        divergence_damping_factor * C123(wgradₕ(divₕ(ᶜ∇²u))) -
        C123(wcurlₕ(C123(curlₕ(ᶜ∇²u))))
@. Yₜ.c.uₕ -= ν₄_vorticity * C12(ᶜ∇⁴u)

---

These changes collectively improve the maintainability and readability of the hyperdiffusion implementation, while ensuring consistent and correct use of model parameters throughout the codebase.

[haakon-e]

This change is part of the following stack:

• rft: clean up hyperdiffusion #4049 ◀

_{Change managed by git-spice.}