Screen for mismatched PDFs

Occasionally the PDFs become mismatched with the paper metadata. Not sure at what point in the pipeline this happens. #7060 is an example.

Could the ingestion pipeline extract text from the PDFs to flag these cases? Suggested by @mbollmann

Start by comparing the mismatched example in issue #7060 and tracing where the ingestion pipeline handles PDFs and paper metadata. Define how extracted text should be compared and how cases should be flagged; done means the pipeline reliably identifies mismatches without disrupting ingestion.

由索引模型根据 Issue 内容生成。