Paper pipeline: structured reading queue with DONE read-state

[Jammy2211]

Closes #35. Third of the three knowledge-board follow-ups — the growth path named at #32 ("my paper management github repo resource").

• Structured queue: section labels become real ## headers (the bare Label: form was one colon-terminated paper title away from misparsing — three such titles exist); format documented in the file header; the stray pre-section title moved into ## SETI.
• Read-state: a consumed paper is never deleted — its line gains a DONE <YYYY-MM-DD> — prefix (Mind-queue precedent), so the queue doubles as the reading history.
• Board: scripts/board.py parses the headers, counts DONE separately → per-section "N waiting · M read" on the Pages board and md; the 📋 file-the-next-paper prompt now ends at "mark the line DONE" instead of deletion. index.md documents the format.
• arXiv ingest (human-gated one-tap): landed on PyAutoMind main (f7a76330) — the nightly Slack digest now ends with a paste-ready Claude prompt appending the surviving papers to ## Strong Lensing; the digest itself still never commits.
• Note: the dev-box's uncommitted queue hand-tidy was accidentally swept into Wiki hygiene: archive: provenance field, alias retirement, link/bib cleanup #36's squash by a git add -A (disclosed there); this PR builds on it.

17 tests green (make validate + board fixtures updated for the header/DONE format). After merge the board refreshes via knowledge_board.yml's push trigger.

🤖 Generated with Claude Code