PDF text extraction moves to pdf-inspector behind a single seam, dropping
markitdown (and with it magika/onnxruntime/pdfminer). Local mode now
prefers our own parser over .zotero-ft-cache, and attachment_priority
makes the PDF-first ordering configurable (#378).

See CHANGELOG.md for the full entry, including the upgrade note: existing
indexed items are not re-extracted merely by upgrading; use
'zotero-mcp update-db --force-rebuild' to move an existing index onto the
improved text.