Add tests for messages with non-standard characters

[chkoss]

Description of Problem:

Currently, there are not enough tests for processing of messages / training examples with a wide range of non-standard characters. Bug with processing such messages can go unnoticed when releasing a new Rasa Open Source version, and will only be caught when trying to release a new Rasa X version. They will then also block the Rasa X release (as happened with Rasa X 0.30.0).

Overview of the Solution:

Add proper emoji testing to Rasa Open Source (e.g. similar to the integration tests in Rasa X here and here). A good list to test on can be found here: https://github.com/minimaxir/big-list-of-naughty-strings/blob/master/blns.json