Refactor for relabeling fields

[BryceStevenWilley]

• Use FormField objects when possible, instead of passing around raw Pike pdf objects
  • be gets font size from DA (PDF spec 12.7.3.2)
• group fields by page (needed to match fields to surrounding text)
• filter out fields without a page, or with a Template page (PDF spec 12.7.6)
• make names longer when reading them (PDF spec 12.7.3.2)
• improve auto_fields_all.py, with proper arguments and everything
  • lets you take all PDF in a folder and either
  • add fields if there aren't any
  • add fields if there aren't any and relabel existing fields
  • scrap existing fields and add fields back with our recognition
  • do nothing: just copy the PDFs

Current at 9a/ff of all forms, will need to re-run ~200 or so that errored out, and then will make another PR adding the stats generator.

Also, Fix #32, was low hanging fruit.

[BryceStevenWilley]

Finished relabeling all 15k PDFs after about 30 hours. It's definitely slow (around 7 seconds per PDF), but I'm not sure how to make it faster, since a lot of it involves the CV stuff. Overall, it looks like it'll take ~40~ (edit: 60) hours or more to run the whole pipeline again. We'll see how it goes.