add BartModelWithDecoderPositionIds base model

[ArneBinder]

This PR adds a variant of the PyTorch BART model with a slight modification where the decoder accepts predefined position ids. If not provided, the original logic is used to create the position ids.

The model is based on the BartModel from Transformers 4.35.0, i.e. https://github.com/huggingface/transformers/blob/v4.35.0/src/transformers/models/bart/modeling_bart.py.

Note: This also contains some minor modifications to make the code mypy (v1.4.1) compliant.

EDIT: See #27.

[codecov-commenter]

Codecov Report

Attention: 18 lines in your changes are missing coverage. Please review.

Comparison is base (38389af) 96.00% compared to head (686dff3) 95.52%.

Files	Patch %	Lines
...dels/base_models/bart_with_decoder_position_ids.py	88.23%	18 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #26      +/-   ##
==========================================
- Coverage   96.00%   95.52%   -0.49%     
==========================================
  Files          32       33       +1     
  Lines        2303     2456     +153     
==========================================
+ Hits         2211     2346     +135     
- Misses         92      110      +18     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.