Make non-compliance eval results graphical in evals/readme.md

[bzorn]

Currently the eval results when you run multiple models and multiple prompt samples is a table that looks like this:

It would be great to also have a bar-chart of non-compliance (where higher is better) that looks like the results presented in the paper:

[bzorn]

We have a script in python to generate mode complex visualizations based on the overview.csv files.