Add training data for one special error case

[lfoppiano]

This PR provides the training data to improve the recognition related to error case #632. I've added in the guidelines a couple of additional sentences related to the author contribution as discussed.

Where I was not sure, I commented directly in the XML files.

[kermitt2]

Hello Luca !
Would it be possible to send me the PDF for these training cases?

[lfoppiano]

This is just one document.
The PDF was linked in the issue: https://www.nature.com/articles/s41598-020-58065-9.pdf

[kermitt2]

Raw header file need to be regenerated after retraining of the segmentation model, because some content has been added in the segmentation file.

I think it would be better here to roll-back the changes in the guidelines because they will conflict will most recent ones done in other place and they are redundant.

[kermitt2]

It might be appropriate to merge this PR with branch feature/data-availability-statement first and not with master, because the changes relatively to training in different branches here make things difficult to follow and update in a consistent manner (the raw files and model in particular).

[lfoppiano]

Thanks for the review! I close this and add these files (without guidelines) directly in the feature/data-availability-statement.

[kermitt2]

closing because it is moved into feature/data-availability-statement