Add misclassified text

[evidencebp]

Misclassified texts are valuable in model improvements.
They are hard to find and might indicate an implementation bug or linguistic expression that is not covered well.

If you find such a case, please open an issue with:

• The text
• The model that should classify it
• An explanation why the text should be classified differently

Extra

• Adding the text to the relevant tests
• Gut feeling why doe the model fail
• A pull request fixing that (without lowering the performance in other cases)