Add four agent memory and planning papers

[estelledc]

Summary

• Continue the study project with a bounded 4-paper round after the agent security/robustness round.
• Fill the next gap: agent inner-loop memory, reflection, long-term user memory, and search-based planning.
• Add study-v2 notes and matching study-review-receipt-v1 receipts for Generative Agents, MemGPT, MemoryBank, and LATS.
• Refresh atlas/note-index and update public scale copy to 1083 papers / 961 projects / 2044 notes.

Evidence Boundary

• All four notes stay STATIC_ANALYSIS / UNVERIFIED.
• No sandbox simulation, MemGPT run, SiliconFriend reproduction, LATS rollout, or benchmark replication was executed.
• Metadata was checked via arXiv API and lr search arxiv.

Verification

• node scripts/quality-gate.mjs for all 4 new notes
• npm run audit:content-contract
• npm run atlas
• npm run audit:counts
• npm run audit:links
• npm run audit:wikilinks
• git ls-files -co --exclude-standard -z | node scripts/audit-public-redlines.mjs --stdin0
• npm run build:strict -- --log /tmp/study-20260715-agent-memory-round-build-clean.log
• STUDY_CHANGED_FROM=origin/main npm run verify:ci