We need to trigger extractors per page even if a barcode repeats, e.g.
for multi-leg IATA BCBPs.

The deduplication logic was apparently always broken and got now fixed
with the new PDF image reference type. This then made things break here
with deduplication becoming too aggressive.