The old from;to CSV applied a free substring replacement to both nom and
prenom indiscriminately - hard to read (no sense of who a rule belongs
to) and unable to express more than one variant per person. Replaced by
one row per doctorant (nom;prenom;autres_noms;autres_prenoms), matched
by exact comparison against a canonical form built per field, with
autres_noms/autres_prenoms accepting a comma-separated list so a
marriage name change and a typo can coexist on the same person.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>