564 chardet performance binary content

[mlissner]

It turns out that r.text makes calls to chardet each time it is called. That's
not great because chardet can be slow and use a lot of memory, particularly
when checking PDFs.

Instead of doing that or checking if things are PDFs all the time, simply use
the binary content instead of the text.

Fixes: #564
Relates to: psf/requests#6250

[albertisfu]

Thanks this works well and seems good to me.

When checking this I noticed a test that needed to be updated after I changed the NDA magic link URL to skip the attachment page, so I added the fix here, hope it's ok.

The weird thing is that tests action passed even though there was that broken test:
https://github.com/freelawproject/juriscraper/actions/runs/3154850047/jobs/5132902032#step:5:2230

[mlissner]

The weird thing is that tests action passed even though there was that broken test:

Huh? Does it return an exit code of zero when you run the python setup.py test command on failing tests?