Fix Feishu PDF upload handling for custom agents

[yaojin3616]

Summary

• route non-image Feishu uploads into an actual LLM processing flow instead of only saving the file and replying with a generic ack
• auto-extract supported office files like PDF into a companion markdown file so the agent can read text immediately
• add regression tests for Feishu PDF upload handling and the synthetic prompt builder

Root cause

Feishu file uploads were saved to workspace/uploads/, but non-image files stopped there. The agent never entered a document-processing path, so PDF uploads were not actually handled after arrival.

Validation

• PYTHONPATH=backend pytest backend/tests/test_feishu_service_api.py backend/tests/test_feishu_file_uploads.py -q
• python3 -m compileall backend/app/api/feishu.py backend/tests/test_feishu_file_uploads.py

Closes #175
Closes #176

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4ceb9c0234

ℹ️ About Codex in GitHub

Codex has been enabled to automatically 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 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Use read_file for non-document uploads

The fallback branch always instructs the model to call read_document for any non-extracted file, but read_document only supports a limited set of extensions (PDF/DOCX/XLSX/PPTX/TXT/MD/JSON/CSV/LOG). For common uploads like .py, .yaml, .xml, .js, etc., this new auto-processing path will immediately hit “Unsupported file format”, so the file flow regresses into a failed first tool call instead of reliably reading content. Please branch the prompt by extension (or direct unknown/text-like files to read_file).

Useful? React with 👍 / 👎.