ORB-4: Merge with Radnext NLP repository
The following issues were addressed [date 23.11.2023]
- fixed OrbGrading description as the SciPost had repeated grade name instead of description
- SciPost venue id is extracted from full name, it is not "SciPost" for all the itmes
- added <purge_submissions_without_review> function to ScipostRawDataset
- purging submissions for ORB dataset is now optional and parametrized
- OpenReview grading fixed when multiple colons ":" are in the grade text
- OpenReview and SciPost boolean decision flag is predicted based on mutliple specific keywords (not only "accept/publish" as before)
- found and covered corner-cases when OpenReview final decision was not in 'Decision' note but in 'Meta_Review'
- optimized export to JSON - instead of dumping whole json-string the JSON Encoder is used to iteratively dump the chunks of data