Support recent OCR-D code

[stweil]

QuiVer benchmarks is currently based on a rather old ocrd/all:2023-03-26.

It would be good to run benchmarks with recent versions of OCR-D, but the code contains assumptions which prevent that:

• It expects models in /usr/local/share.
• Others?

[stweil]

@mweidling, are there plans to update the code to support latest OCR-D? Or did you stop the work on QuiVer benchmarks?

[mweidling]

@stweil Yes, there is. I'm currently working on it in the update-ocrd_all branch.