[DONT MERGE] simple text classification

[ArneBinder]

Quickstart:

1. click on the green button Use this template -> enable Include all branches and choose a repository name, e.g. my_cool_project -> click on Create repository from template
2. clone the new repo, cd into it and call git switch simple_text_classification
3. follow the remaining steps of the main Quickstart instructions (without the first step to clone the repo)
4. setup Weights & Biases: create an account, if you don't already have one, and generate an API key from here
5. start example training, e.g. on imdb:
   python train.py experiment=imdb (optionally add the parameter trainer.gpus=1 to train on a GPU)
6. adapt or create any dataset, taskmodule and / or model for your needs

Notes:

• Maybe have a look at the diff of this PR to see what was needed to get this use-case working.
• Per coincidence the imdb dataset consists of documents of a type that matches the requirement for SimpleTransformerTextClassificationTaskModule, i.e. they have a text and a label annotation field. If this is not the case, pytorch_ie.Datasets provide a method cast_document_type that allows to re-map or hide any fields of the contained documents.