Import of large PDFs

[floschne]

This fixes the import of large PDFs by fixed page-based chunking and improves PDF import in general by using Docling.

STRATEGY:

0. check if PDF needs to be chunked, i.e., if it has more than N (per default 5) pages.

YES:

1. Chunk the PDF
2. stop prepro for cargo (not the whole PPJ!)
3. create a new PPJ from the chunks

NO:

1. continue with extracting content as HTML including images from PDF via Docling through RayModelService

Open Questions:

• how to properly link the chunks concerning the page order and SDoc links to navigate in the UI?
• can we maybe have some sort of Parent SDoc (no Adoc!) that links the chunk sdocs?

TODOs:

• write an end2end test! I already uploaded a good example file at ltdata. Since multiple PPJs are spawned from the chunks and their images, one naive strategy would be to upload the file, set a max time limit (about 10 mins on GPU), poll the PPJs, assert that all PPJs are done within the time limit.
• improved chunking strategies (semantic, structured, etc)
• improve throughput by batched docling conversion (https://docling-project.github.io/docling/examples/batch_convert/)