Dynamic discovery of OCR languages

[apyrgio]

One of the main contributing factors to the size of our container image is the OCR models for all the languages that we bundle with our application:

# du -hd 1 / | sort -h
...
1.8G    /usr
1.8G    /

# du -hd 1 /usr | sort -h
...
717.7M  /usr/lib
1.1G    /usr/share
1.8G    /usr

# du -hd 1 /usr/share/ | sort -h
...
1015.6M /usr/share/tessdata   <--- OCR models
1.1G    /usr/share/

We don't expect that we will have any user that will need all of the Tesseract models simultaneously, so we need to find something better here. An interesting idea would be to dynamically fetch the models necessary for the conversion (e.g., under Dangerzone's data directory), and mount them to the container under /usr/share/tessdata.

There are two issues with this approach:

1. The canonical repo would be GitHub, which may reject requests due to rate-limiting.
2. This is something that cannot work on airgapped installations, so we would need instructions for those.