Fix missing coordinates in paragraphs continuation

[lfoppiano]

When the paragraph continues after interruption (e.g., reference callout), the coordinates are lost:

This PR solves this issue.

This PR also adds a small modification in the frontend so that the paragraph coordinates are extracted if "add coordinates" is selected and "segment sentence" is not selected.

[coveralls]

coverage: 39.906% (+0.01%) from 39.892%
when pulling 0d7913d on bugfix/paragraph-coords
into cbc77d5 on master.

[kermitt2]

I didn't see the problem in the previous PR, sorry !

[lfoppiano]

Neither did I when I was developing it. The structure viewer app (https://structure-vision.streamlit.app/) is quite helpful in validating the stream order of PDF extraction.

[Darrshan-Sankar]

Still there are few paragraphs un-annotated it seems. I checked in a PDF. Any fixes?

[lfoppiano]

Oulx you please share an example?

[Darrshan-Sankar]

Please check with these articles:

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10510434 ------> Article page no:9
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10042417 --------> Article page no:8

And similarly many more articles

[lfoppiano]

Those issues are not related with this PR. Here the issue is that part of the text is misclassified as figure.

I've referenced your comment in a separate issue. This will likely be solved, or, at least, mitigated by #963 (WIP).

[Darrshan-Sankar]

@lfoppiano Thanks looking forward for the fix