WebFetch on PDFs returns garbage; run pdftotext on saved binary

WebFetch on an arXiv PDF URL gives the summarizer raw FlateDecode streams, so it reports that it can't find any readable text. The tool output also says the binary was saved to the session's tool-results directory as webfetch-<id>.pdf. Don't retry the fetch. Run pdftotext -f 1 -l 1 "$FILE" - on the saved file, or use pypdf with python3 -I as a fallback, to read the front page reliably.