Add converted markdown for new sources

[aadehamid]

Summary

Automated daily source conversion run (scripts/convert_pdfs.py).

This run produced 1 new converted markdown file plus its metadata sidecar:

• references/papers/llm-inference-optimization/gptcache.md — GPTCache: Semantic cache for LLMs (converted from https://github.com/zilliztech/gptcache)
• references/papers/llm-inference-optimization/gptcache.meta.json — metadata sidecar

Notes

• All other queued sources were already converted; only their .meta.json fetched_at timestamps changed, so those pre-existing files are intentionally not included in this PR.
• A few arXiv /abs/ URLs returned HTML instead of a PDF and were skipped by the converter; two Medium articles returned HTTP 403. None of these produced new markdown.

Conversation: https://app.warp.dev/conversation/e4571dcf-4c39-45ad-90d7-8d2adaccef3a
Run: https://oz.warp.dev/runs/019ece5d-2013-71b1-981c-6b97d175c237

This PR was generated with Oz.

[github-actions]

[daily-source-conversion-reviewer] Review complete at 2026-06-16 03:04 UTC
Files reviewed: 1 | Subject areas: llm-inference-optimization

Check	Result
(a) Not empty	✅ all passed
(b) Readable markdown	❌ 1 failed: references/papers/llm-inference-optimization/gptcache.md
(c) meta.json valid	✅ all passed
(d) Filename consistency	✅ all passed
(e) No duplicate URLs	✅ all passed
(f) No scope creep	✅ all passed

Outcome: ❌ Changes requested — PR NOT merged

[github-actions]

Found issues in 1 file(s):

• references/papers/llm-inference-optimization/gptcache.md — (b) Readable markdown failed: file starts with GitHub navigation/boilerplate capture instead of clean converted source notes.