Add Paper2Any handoff bundles

[ZimoLiao]

Summary

• add managed Paper2Any runtime support: scholaraio paper2any setup, status, capabilities, serve, and prepare
• keep the full upstream Paper2Any checkout outside ScholarAIO source under data/runtime/extensions/paper2any/Paper2Any, with local overrides in config.local.yaml
• install/check the isolated Paper2Any Python venv plus standalone CLI, FastAPI, and frontend export dependencies so non-developer agent users do not configure those by hand
• make prepare generate runnable bundles under workspace/_system/paper2any/, stage outputs inside the Paper2Any checkout when upstream requires it, then copy real artifacts back to the bundle
• cover every current upstream standalone CLI script: figure, ppt, ppt-classic, pdf2ppt, image2ppt, ppt2polish, poster, and video
• expose upstream API-only workflows through paper2any capabilities and paper2any serve, including DrawIO, rebuttal, citation, image playground, mindmap, file, and KB routes
• redact upstream API-key log output from generated runners while preserving failing exit codes
• update the Paper2Any skill, CLI docs, setup check, and tests so agents can discover the correct CLI-vs-API path

Real Paper2Any Smoke

• paper2any status passes against the real OpenDCAI/Paper2Any checkout at data/runtime/extensions/paper2any/Paper2Any
• paper2any capabilities lists both prepare tasks and API-only workflows
• paper2any serve --port 9998 starts the real upstream FastAPI backend; /health returns {"status":"ok"}
• regenerated all 8 standalone bundles from the PR branch and inspected each manifest/script mapping
• figure: ran the real upstream figure CLI on a ScholarAIO paper and produced PNG/SVG route-diagram artifacts
• ppt: ran the real upstream frontend editable PPT CLI and produced frontend_slides.json, frontend_theme.json, frontend_summary.json, and paper2ppt_frontend_editable.pptx
• pdf2ppt: ran the real upstream PDF-to-PPT path with OCR and produced a 30-slide editable PPTX with source-page images plus OCR text boxes
• image2ppt: ran the real upstream image-to-PPT CLI from a --source-file bundle and produced pdf2ppt_qwenvl_output.pptx; inspected it with scholaraio document inspect and viewed the embedded slide image
• ppt-classic and poster: runnable bundles are generated; direct DeepSeek smoke exposed upstream model-name/image-model assumptions, so these paths require an upstream-supported model configuration for full artifact generation

Verification

• python -m ruff check scholaraio tests
• python -m ruff format --check scholaraio tests
• mypy scholaraio/
• python -m pytest -q -p no:cacheprovider -> 1343 passed, 3 skipped
• python -m mkdocs build --strict
• git diff --check
• secret scan against current API-key environment values -> none in diff
• PAPER2ANY_ROOT=/home/lzmo/repos/personal/scholaraio/data/runtime/extensions/paper2any/Paper2Any python -m scholaraio.cli paper2any status
• SCHOLARAIO_CONFIG=/home/lzmo/repos/personal/scholaraio/config.yaml PAPER2ANY_ROOT=/home/lzmo/repos/personal/scholaraio/data/runtime/extensions/paper2any/Paper2Any python -m scholaraio.cli setup check
• curl -fsS http://127.0.0.1:9998/health during paper2any serve smoke

Closes #64

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 152a7d0e19

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid inlining markdown into a single CLI argument

When input_kind == "markdown", the generated runner reads the entire paper.md into PAPER2ANY_INPUT and passes it as one --input argument. For large markdown files this exceeds the OS ARG_MAX limit and the script fails before Paper2Any starts (e.g., Argument list too long), so valid large papers cannot be processed. Please switch the markdown path to a file-based handoff (or another streaming mechanism) instead of embedding full content in argv.

Useful? React with 👍 / 👎.

[ZimoLiao]

Fixed in bb07f70. Markdown handoff no longer reads paper.md into a shell variable or passes it through exec argv; the generated runner now reads the file inside a Python wrapper and runs the upstream Paper2Any module in-process, while preserving passthrough args such as --api-key. Added a regression smoke with a fake Paper2Any checkout.

[ZimoLiao]

这个 Paper2Any issue 已经在新的干净实现里重新处理：#92。

#92 不再沿用这个 PR 的 handoff-bundle 方案，而是改成轻量 MCP sidecar + 外部 OpenDCAI/Paper2Any runtime extension：

• ScholarAIO 只保留配置、MCP sidecar、CLI、skill 和文档；不 vendoring Paper2Any 源码或依赖。
• Paper2Any checkout 放在 data/runtime/extensions/paper2any/Paper2Any。
• 新增 scholaraio paper2any setup、mcp-serve、backend-serve、status/tools/call。
• 已实际启动真实 Paper2Any backend，并通过 ScholarAIO MCP sidecar 跑过真实 service matrix。

因此这个旧 PR 先关闭，后续请看 #92。