Refactors to improve code clarity and make changing parameterisations easier

[jagoosw]

I think there is still some room to make this more flexible, for example changing nanophytoplankton and diatoms to just phytoplankton which is a NamedTuple of phytoplankton classes, but then we need functions like ammonia_uptake(all_the_phytoplanktons). This would be useful for experimentation in the future with different things but I am not going to implement here as we just want a working version, and then when people want to experiment that can make these kind of changes.

To do:

• docstrings
• show methods
• check for any more deviations from Aumount 2015
• Fix negative tracer scaling
• comment all parameters units
• documentation

[jagoosw]

So I had todo some horrible things to get this to work on GPU because it seems that _apply_iterate is not implemented in CUDA C or CUDA.jl or somewhere along the line for very long tuples, so this code also introduces a hack which I will hopefully implement in Oceananigans.

[jagoosw]

I think there is a whole new level of abstraction we could do by changing the model to be discrete form, which will likely make this run more efficiently, as well as facilitate grouping the phytoplankton etc. as discussed above.