I'd like to avoid patching in dependencies whenever possible. From what I understand here, tesseract running locally relies on certain dependencies only available in Ubuntu 20.04, the last time the local code was compiled. I think the better move here- and which actually matches what production does, is to make tesseract self contained when running locally. I have a branch that does this here:
https://github.com/MuckRock/documentcloud/tree/local_ocr_fix
I feel that this is better than patching the local.yml file as it is more explicit about what dependencies are exactly where they live and makes it more self contained. I have verified that OCR works locally with this branch

I think these two approaches are actually similar - one is doing a multistage docker build in order to grab the necessary libraries from the Ubuntu 20.04 image, while the other is just checking those libraries directly into git.