Add system regression tests

[theferrit32]

When we make changes to prompting or how fields are acquired we should run an end-to-end test to see if any functionality has regressed, using a set of papers and the prior set of results returned for them.

This would also help us see when functionality has improved. We should also just go ahead and store the PDFs and results for each in a bucket. These are open access in pubmed but we should go with a private bucket for storing these files because we will probably want to add other ones with more restrictive licenses later.

pmid:36704923
pmid:28824526
pmid:35227461
pmid:33193064
pmid:32733554
pmid:32122354
pmid:35712613
pmid:32949214
pmid:31296563
pmid:34200226

[theferrit32]

From 2026-01-08 meeting we should also see if we can take some of the manually curated results from the evagg evaluation to test new code against to identify regressions or performance improvements.

i.e. if evagg returned a particular value for a particular field we should be able to do the same. Or if evagg returned an incorrect value for a field, or missing field, we should identify when we return a different value and whether the new value is correct.