Reduce amount of model runs in (text) tutorials

[loostrum]

In e.g. the lime text tutorial, DIANNA is run separately for each class. This is unnecessary, wwe can also run it once and request heatmaps for multiple classes. In my opinion this makes the tutorials more readable as well. As the text model is quite slow, the multiple runs take a lot of time too.
I have not checkd if something similar is happening the other tutorials.