Dropping ambiguous publication entries from literature

[DSuveges]

As a user I want unambiguous pmid -> pmcid mappings in literature because now multiple pmcid is linked with a single pmid leading to confusing behaviour. (for more context, see #3053 and #2970)

Although the bug is originated upstream to OT, we should take care of this ambiguity before publicly releasing the post-etl literature data. By nature, this issue only affects full text articles.

Tasks

• [ ] Drop all entries where the pmid maps to multiple pmcid

Acceptance tests

• [ ] Ensure the following query return zero documents:

(
    matches
    .groupby('pmid')
    .agg(
        f.collect_set(f.col('pmcid')).alias('pmcids')
    )
    .filter(
        f.size(f.col('pmcids')) > 1
    )
    .count()
)

[mbdebian]

@remo87 , would you mind checking with @DSuveges on this one? Thanks!

[DSuveges]

The problem still exists, as the above quirey flags the following pmids:

+--------+-------------------------+
|pmid    |pmcids                   |
+--------+-------------------------+
|32790207|[PMC10081512, PMC7285927]|
|33376052|[PMC7709584, PMC7983453] |
|31745814|[PMC7574644, PMC6940410] |
+--------+-------------------------+

That's all. Out of the millions of pmids. If there's no way to find out which is the "real" pmcid, we can just drop them.