Pre-allocate grid-point pressure values and use consistent naming

[white-alistair]

This is a left over from this comment on #82

[milankl]

On this note, for the shallow water model, I've decided to use the interface displacement η as prognostic variable, so the instantaneous layer thickness h = η + H, whereby H is the thickness at rest and H = H₀ - orography and H₀ a global constant, something like 8-12km probably. Having said that, I don't want to redefine the variable names in PrognosticVariables for the different models, so I'll be using pres (or even pres_log if we decide to redefine it) for η. This is probably an argument to keep this variable name quite generic (I vote for pres, as this has a generic meaning, whether you think about logarithm of surface pressure or interface displacement).

[white-alistair]

@milankl picking this up again. How about the following approach:

1. Pre-allocate a new grid variable pres_exp_grid in the GridVariables struct (the name is potentially misleading, but I think all of the options are suboptimal in their own way, so I'd be happy to go with your preference from the previous comment).
2. In tendencies_parametrizations.jl, do pres_exp_grid = exp.(pres_grid), i.e. before any of the parametrizations are calculated, since this quantity is used by several of them. I'm not 100% this is the most logical place to do this, but as far as I know we only use this in the parametrizations.

Let me know what you think!

[milankl]

Is pres_grid (i.e. the logarithm of surface pressure on the grid) even used for the tendencies? If not, we could just have pres_grid which is first transformed from spectral pres and then calculate the exponent in-place? Then pres would always be the logarithm of surface pressure in spectral space and pres_grid the surface pressure (hPa) in grid-point space.

[milankl]

I agree that we should have one place where the exp is calculated as this is expensive (usually taking several times more cycles than +,-,*,/ etc) but maybe we don't have to allocate another array but can do it in-place like

for i in eachindex(pres_grid)
    pres_grid[i] = exp(pres_grid[i])
end

[white-alistair]

As far as I can see it's only used for the physics, so yeah I think this works. I'll just comment it clearly so newcomers aren't confused :)