I suspect one of the prime reasons to be the involvement of all the nested and otherwise invisible borders, I've ran into some similar problems in the past and a viable workaround for me is to use tools stronger in extracting text with positional information like pdfplumber. Extracting tables right away appears to be difficult in this case and having a two-fold approach of extracting (not tabular but still correct and well-spaced) text first and then some additional manual parsing on top via tools like regex or parse could be a good way forward.