Integration test: Document reference

[christianabbet]

Close #96 (comment)

Description

Add tests to verify the output coherence of the pipeline. This complements .github/workflows/pipeline_run.yml, which validates that the pipeline runs successfully but does not check the consistency of its output.

To perform these tests, a PDF is constructed from single pages covering all supported classes using the script src/scripts/create_reference_document.py. The resulting reference_document.pdf is fed through the pipeline. The tests then verify that:

• Page count matches the number and index of detected pages
• Entities fall within the document page range
• Document page classification is coherent with detected entities (classes match across input and output)