Perform OCR and extract Word documents to avoid losing about 4.5% of documents

[vitawasalreadytaken]

We're currently disregarding scanned documents which don't contain the text in digital form. The latest counts from the preprocessing pipeline are:

Missing content for 2366 documents (4.5%):
document_source
fedlex          1016
openparldata    1350

Instead of dropping them, we should attempt to extract the text using OCR.

[vitawasalreadytaken]

Actually, about 880 Fedlex documents are MS Word documents. The problem there isn't OCR but that we're unable to process Word at the moment.

09:25:57.069 | WARNING | Task run 'extract_document_content-3ff' - Ignoring 903 non-PDF files (86.7%):
stored_file_mime_type
application/msword                                                         515
application/vnd.openxmlformats-officedocument.wordprocessingml.document    371
text/html                                                                    6
application/vnd.openxmlformats-officedocument.spreadsheetml.sheet            6
application/zip                                                              3
application/x-empty                                                          2
Name: count, dtype: int64