implement serialize_brat() and add test for that and load_brat()

[ArneBinder]

• insignificant refactor of load_brat, semantic is still the same (just outsourced the loop and variable renaming)
• minor improvement to hf_datasets/brat.py: if data_dir contains subfolders, take these as split names (if subdirectory_mapping is not already specified)
• re-introduces functionality to create train test splits directly in load_brat when parameter train_test_split: Dict[str,Any] is available and the loaded data has only one split. The parameter train_test_split gets passed to loaded_split.train_test_split as kwargs.
• add serialize_brat: serialize a corpus of documents (i.e. Dict[str, List[Document]]) to strings or write into folder(s)
• add test for load_brat
• add tests
  • load_brat
  • serialize_brat - in memory
  • serialize_brat - write into folder
  • serialize_brat - full cycle, i.e load -> serialize -> load

[ArneBinder]

@ChristophAlt just rebased onto main, could you have a look at this, when you have a bit of time?