Avoid refextract storing PDF on HDD - prefer working in RAM

[will-afs]

[will-afs]

Tried to rewrite refextract relevant functions.
But it seems refextract relies on the magic library which requires itself using PDFs accessible from filesystem
Also tried to use refextract.extract_references_from_string(), but does not work neither

So if we want to continue using Docker container and AWS Lambda, maybe changing of library for extracting references would be a better option.