Add function to plot Impf variability of calibration

[timschmi95]

Changes proposed in this PR:

• add a function which plots the impact function variability given a certain percentage difference in the cost function values. Additionaly has the option to plot the distribution of hazard intensity values at the selected exposure points

As discussed at the retreat. For me this function was quite useful to get a "feel" for the data. In some cases, it highlighted completely different impact functions with an almost equal cost-function value. Let me know if you think such a function is useful for users in general.

The two added plots and description are in the Section "Analyze the Calibration" of the tutorial Jupyter notebook.

PR Author Checklist

• Read the Contribution Guide
• Correct target branch selected (if unsure, select develop)
• Descriptive pull request title added
• Source branch up-to-date with target branch
• Documentation updated
• Tests updated
• Tests passing
• No new linter issues
• Changelog updated

PR Reviewer Checklist

• Read the Contribution Guide
• CLIMADA Reviewer Checklist passed
• Tests passing
• No new linter issues

[leonie-villiger]

Thanks for adding the method .plot_impf_variability() to the OutputEvaluator! I believe this nice and informative visualisation is very helpful when calibrating impact functions. My comments:

• code comments could be a bit more specific (see suggestions in base.py)
• don't fully understand the reasoning to show the intermediate step of getting the haz_vals before plotting (see climada_util_calibrate.ipynb)
• several trailing whitespace Linter issues (VS can remove the automatically, https://linux.how2shout.com/remove-trailing-spaces-automatically-in-visual-code-studio/)
• complete the PR checklist (just now saw that the PR is only a draft...uuuups. I can give a final review once you're fully ready).

[timschmi95]

Thanks @leonie-villiger for the review, especially the automatic whitespace removal tip! :)
Since this is a pull request within a hanging pull request I didn't update the documentation or Changelog, but marked it as done.

@peanutfun I managed to get the histogram much faster by using the .data attribute. Do you want to have a look over it before I merge it? The failing unit test are not related to this specific pull request, and I didn't add any tests for the new plotting function for now (as with the other plotting function in the calibration module). Is that all fine?

[peanutfun]

@timschmi95 Very helpful contribution, thanks a lot for that! My main concerns are how to deal with multiple impact functions within the impact function set and with the handling of plot parameters.

The failing tests are fine, we will fix this in the "main" PR.

[timschmi95]

@peanutfun Thanks for the review, let me know if you think the added changes are suitable. Currently, the function will only return the last axis object in case there are multiple impact functions, but I think this shouldn't be an issue?