Draft: Resolve "Implement npzd model"

Closes #2

We would like to have a modular framework that enables auto-differentiable simulations of advection and diffusion, but also biogeochemical processes. While the former already works well and many numerical schemes are implemented (just like in the original GOTM), no biogeochemical models were implemented. Now, there is a light model and a simple NPZD model. This model which has only four types of tracers is one of the simplest examples of a biogeochemical model. Combined, the light and NPZD model can reproduce the fortran GOTM output with a high accuracy, taking as additional input only the incident solar radiation.

Issues that still have to be addressed:

• Implement do_ppdd method for NPZD model (needed for sink/source term flux matrix schemes, i.,e. Patankar schemes)
• Implement further schemes (currently only Euler first order)
• Figure out how best to combine models, e.,g. an NPZD-type model and a passive tracer
• Clean up all model inputs: Are they individual inputs or elements of the dynamic_inputs tensordict, which dimensions do they have?
• Add a model that has sediment / surface steps, implement support for those
• Add ECOSMO-type model
• Add unittests
• Add demonstration notebooks
• Write docstrings