Generate syms.txt

Using the work done in https://gitlab.com/teklia/htr/pylaia_scripts/-/blob/main/pylaia_scripts/convert_to_pylaia.py:

• Port the Syms class
  • remove unrelated functions like set_phase or print_unk_chars
  • create a new staticmethod from_lines(lines: List[str]) that will create a new Syms object and fill its _syms attribute. This method will add every character (not spaces) from the given line transcriptions to _syms.
  • Remove unnecessary code from char_replace as the Partition and read_only features will not be needed here.

The plan is to use it in generate_dataset.py.

• When you have the valid polygons, we will iterate a first time on the training dataset to retrieve every line's transcription (keep the first transcription found if there are any, skip the line otherwise).
• Then we create the syms instance using syms = Syms.from_lines(lines) and save it to disk using syms.to_file(syms.txt).
• Then we can go over the lines again and call the existing process_line method but with the tokenized transcription text using syms.process_line(transcription["text"].