sync: fix PDF candidate selection to consider all PDFs and prefer main.pdf/paper.pdf at root

Fix #33: restore dc:creator XMP field to real authors after setCreator()

Fix #32: set PDF Creator field to iTagPDF (www.itagpdf.com)

Fix #27: generate one LI per list item, not per line

fix: downscale large images before Tesseract OCR to prevent hangs

fix: replace executor on cancel to unblock new pipeline runs

debug: add trace prints to upload route to find hang point

fix: log google auth error reason; export pipeline env vars in start.sh

Add itagpdf.com to CORS allowed origins

fix: commit pdftagger.jar via LFS (was excluded by .gitignore)

Sync latest backend and pipeline code

fix: remove pipeline cancellation so concurrent users queue instead of cancel

fix: mvn shade outputs pdftagger.jar to project root, no cp needed

exclude large notebook from repo

add pdf-script as plain directory

remove pdf-script submodule reference

build pdftagger.jar from source using maven

add pdftagger.jar via LFS

fix: use default-jdk (openjdk-17 not available in Debian trixie)

Initial deployment