It's never going to be all knowledge, since a lot of stuff is just lost or never recorded. A ton of stuff (like this thread) are probably low on the priority list for recording as well. But the closest you'd probably get to a full catalog of human knowledge (at last text based) are the huge data sets of nearly all text data on the internet used for training LLMs. I wouldn't be surprised if there are ones soon that include video and pictures as well, since newer AI models are starting to be able to interpret those too.
I believe this is one of those data sets: https://github.com/yaodongC/awesome-instruction-dataset
Edit: here's a big data set used for a lot of gpt3 https://commoncrawl.org/