Unit tests for extraction command

We need a better test suite for extraction command. We need to create the SQLite database as a fixture during tests using arkindex-export. We need a clear schema of the data inside this DB. We can start with something simple:

• root dataset folder with train/val/test subfolders
• 2 pages per split, with 5 lines on each pages
• 1 manual transcription and 1 transcription with a dedicated worker run source, on the pages and the lines
• 2 entities per transcription on pages and one/two of the lines on each page