Add PaddleOCR and Refactor Text Extraction

[karanataryn]

• Refactors the Text Extraction interface to contain both PDFMiner and OCR, allowing them to act as part of the pipeline per page or supplement text to the whole document.
• Keeps the old OCR implementation under the per_element_ocr flag.
• Adds PaddleOCR, Tesseract, EasyOCR, and Legacy (Easy + Tesseract) as OCR options.
• Refactors code to make it cleaner and changes necessary documentation.

[alexaryn]

Looks generally OK to me. I'd try to avoid calling PdfMiner an OCR model in the code, just to avoid confusion, but I get that it serves a similar purpose.

[alexaryn]

Interesting dilemma here. Changing to page-at-a-time makes it harder to cache PdfMiner output. Yet, PdfMiner still represents a significant chunk of work that would benefit from caching. Is there a way to stream PdfMiner output into the cache and stream it out? The only approach that comes to mind is disk spooling, which I expect to be faster than PdfMiner itself.

[karanataryn]

Agree that there isn't a clear solution here. I think we should wait on this until after this PR, and decide on an approach for caching for both OCR and PDFMiner.

[karanataryn]

Looks generally OK to me. I'd try to avoid calling PdfMiner an OCR model in the code, just to avoid confusion, but I get that is serves a similar purpose.

It's a Text Extractor not an OCR Model in the code. Only the OCR Models are called as the latter.

[bsowell]

I think I'm happy with this. Just want to understand what we are setting the defaults to.

[bsowell]

One minor comment, but otherwise I think this looks okay.

[alexaryn]

A number of comments. I read it in detail.