Automatically change parameter and return float types

[jagoosw]

This PR removes hard coded float types in function returns, and sets up most (not worth it to do the sugar kelp model because it has a billion parameters) model setup functions to automatically set the correct float type, or at least be able to set in a simple way, e.g.:

grid = RectilinearGrid(Float32; ...)
bgc = PISCES(; grid) # now has the correct FT
cc = CarbonChemistry(Float32)

cc: @ali-ramadhan @glwagner

Closes #218

[glwagner]

That looks nice, two comments:

1. I prefer bgc = PISCES(grid). The interface is a bit split whether grid should be an arg vs kwarg but I think we have been gradually drifting towards grid as positional arg (the big change is for models, which will have to come later).

2. There's a discussion about making it easier to change number type here: How can we help users change number type more easily? CliMA/Oceananigans.jl#3800

[glwagner]

(not worth it to do the sugar kelp model because it has a billion parameters)

I don't follow this statement --- what do you mean?

For situations with large numbers of parameters, you can put them in an Array (like TabluatedAlbedo for ClimaOcean):

https://github.com/CliMA/ClimaOcean.jl/blob/fc175f6c6b3e095bd460d10d323e4361006a5008/src/OceanSeaIceModels/CrossRealmFluxes/tabulated_albedo.jl#L90

Another possibility is to convert at runtime, eg with something like

const parameter = 1.23

f(a::FT) = a + convert(FT, parameter)