pdf parsing in python is just good as long as youre using it but when someone uploads a 700 page file the worker runs out of memory so splitting it into 50 to 100 page chunks with docling or marker on celery gets you past that but tables get cut at the boundaries and now theres a queue, retries and memory limits to babysit the other route is paying for something like llamaparse or the paid parsers has anyone moved either way and was it worth it? like what did you do inside your tool/app when it started scaling