Update function evaluator

[timmens]

In this PR, I update the code corresponding to the function evaluator.

• Add typing to function_evaluator.py module
• Replace all occurrences of non-numerical discrete labels. (We will not support them.)
  For example, we test the lookup with characters ["good", "bad"], which we do not want to support.
• Replace all occurrences of numerical labels that are not themselves indices (i.e. [0, 1, ..., endpoint] is okay, but [1, 2] is not)
• Rewrite label translator to only return identity function (works on the assumption that all discrete labels that are processed internally are their own indices)
• Throw error when processing a model with a discrete grid that does not correspond to indices
• Add explanation notebook on function evaluator
  • Write motivation presenting the high-level function
  • Write technical section on behavior of small functions
  • Add task to run explanation notebooks on CI (prevent misaligned of LCM codebase and these notebooks)
• Compare map_coordinates to other interpolation tools, e.g., RegularGridInterpolator
  • It turns out that the RegularGridInterpolator implements a linear extrapolation feature. However, it is prohibitively slower compared to map_coordinates. Luckily, we can adjust map_coordinates to also perform linear extrapolation. See notes below.

Note

1. This PR introduces a new error that will be thrown during the model processing. It was already discouraged to use discrete grids that are not indices (since other grids were not implemented and would throw an uninformative error). Now, an informative error is thrown if the user tries to use a grid that does not coincide with indices.
2. During work on this PR, Janos and I discussed options to implement arbitrary discrete grids. We settled on a solution that is described in ENH: Support arbitrary discrete integer grids #82. We do not implement it here, because it will be easier to implement once the model processing has been refactored.
3. While I found a very nice solution to add extrapolation to our function evaluator, we will not tackle any extrapolation issues in this PR since we have to wait for a response from a JAX maintainer; see issue ENH: Add linear extrapolation to map_coordinates  #83 for details.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[hmgaudecker]

Looks great! There may be room for language improvement, but mainly about docs. I already changed a little bit.

[hmgaudecker]

Excellent, thanks!