GPT OCR followup

[danvk]

Addresses most of #147; follow-up to #146

• Moves OCR cleaning into ingest.py
• More aggressive removal of "Negative No. ..." lines
• Strip/break interior whitespace
• Break up numbered lists on a single line
• Strip various low-information stamps:
  • F. S. LINCOLN
  • NEW YORK PUBLIC LIBRARY
  • President Borough of Manhatta/en
• Run OCR cleaner on old site OCR, too. (Strips negative numbers, credit line imperative, etc.)

These affected the OCR shootout results, so I went ahead and updated those, too.