Fix/1209 tweak xycut ordering output

[christinestraub]

Closes #1209.

Summary

• add swapped xycut sorting
• update xycut sorting evaluation script

PDFs:

• sbaa031.073.pdf
• multi-column-2p.pdf
• 11723901.pdf

Testing

elements = partition_pdf("sbaa031.073.pdf", strategy="hi_res")
print("\n\n".join([str(el) for el in elements]))

Evaluation

PYTHONPATH=. python examples/custom-layout-order/evaluate_xy_cut_sorting.py sbaa031.073.pdf hi_res xycut_only

[Klaijan]

Majority of the fixture tests improve with this tweak. Both versions get confused when reading tables or mixed of single paragraph and figures or tables in one page. With table reading, I feel like the previous version can capture better reading order, but not significantly better.

For reference, I have my note here:

test_unstructured/partition/utils/test_xycut.py meh
test_unstructured_ingest/expected-structured-output/azure/IRS-form-1987.pdf.json worse
test_unstructured_ingest/expected-structured-output/azure/IRS-form-1987.png.json worse (better ordering within its section, but worse overall)
test_unstructured_ingest/expected-structured-output/biomed-api/65/11/main.PMC6312790.pdf.json better
test_unstructured_ingest/expected-structured-output/biomed-api/75/29/main.PMC6312793.pdf.json better
test_unstructured_ingest/expected-structured-output/biomed-path/07/07/sbaa031.073.PMC7234218.pdf.json better
test_unstructured_ingest/expected-structured-output/pdf-fast-reprocess/biomed-api/65/11/main.PMC6312790.pdf.json better (meh but will give benefit of the doubt)
test_unstructured_ingest/expected-structured-output/pdf-fast-reprocess/biomed-api/75/29/main.PMC6312793.pdf.json better
test_unstructured_ingest/expected-structured-output/pdf-fast-reprocess/biomed-path/07/07/sbaa031.073.PMC7234218.pdf.json better
test_unstructured_ingest/expected-structured-output/pdf-fast-reprocess/s3/small-pdf-set/2023-Jan-economic-outlook.pdf.json both ok
test_unstructured_ingest/expected-structured-output/pdf-fast-reprocess/s3/small-pdf-set/Silent-Giant-(1).pdf.json better
test_unstructured_ingest/expected-structured-output/pdf-fast-reprocess/s3/small-pdf-set/recalibrating-risk-report.pdf.json better (both have trouble reading tables, but the new code did worse at reading paragraphs)
test_unstructured_ingest/expected-structured-output/s3/small-pdf-set/2023-Jan-economic-outlook.pdf.json both ok (not very good at reading table still)
test_unstructured_ingest/expected-structured-output/s3/small-pdf-set/Silent-Giant-(1).pdf.json better
test_unstructured_ingest/expected-structured-output/s3/small-pdf-set/recalibrating-risk-report.pdf.json better