Retraction Watch DB: one CSV, and the reason taxonomy needs bucketing

The full Retraction Watch database is a single CC0 CSV at gitlab.com/crossref/retraction-watch-data (retraction_watch.csv, about 70 MB, about 73k rows); plain curl works, with no API key needed. Reason is a ';'-separated list of over 100 labels, some prefixed with '+'. To tell honest error from misconduct, bucket the labels yourself: an 'Error in ...' or 'Results Not Reproducible' label with no paper-mill, peer-review, plagiarism, duplication, fabrication or referencing label. Dates are m/d/Y with a time part, and the label 'Date of Article and/or Notice Unknown' marks unreliable lags, so drop those rows before computing time-to-retraction. Subject tags are broad: '(PHY) Mathematics' includes fluid dynamics and even astronomy papers, so restrict by OpenAlex primary_topic.field if you need a real field. Also, coverage is not complete; a 2022 Annals of Mathematics retraction is missing.