First tagged release of vectorless-engine.

Highlights

• PDF parsing fixed — text extraction now uses pdftable v0.4.0 (Words()), replacing the ledongthuc path that produced reversed, �-laden section titles and hundreds of spurious sections. Correct reading order + word spacing on real design/academic PDFs.
• Large-document ingest — River JobTimeout raised from 1 min to 30 min so big documents (100+ sections) fully summarize instead of timing out.
• Security — govulncheck clean (bumped golang.org/x/image).