[ump45nose]

Summary

Fixes #7436.

Uploading a zip that only contains PDFs returned an empty split list with no server logs, while uploading the same PDF alone worked.

Root cause: PdfSplitHandle.handle read the file with UploadedFile.chunks(). Zip members are zipfile.ZipExtFile, which has no chunks(), so the call raised AttributeError. ZipSplitHandle then swallowed every exception with except Exception: pass, so PDF members were skipped silently.

Change:

• Write the PDF temp file from get_buffer(file) (same path Doc/HTML/XLSX already use), which works for both UploadedFile and ZipExtFile.
• Log zip-member split failures with max_kb.warning instead of silent pass.

Related prior attempt: #7448 (closed unmerged against an older impl/text/ layout).

Test plan

• Local smoke of the temp-file write path with a raw buffer
• Added apps/common/handle/impl/tests/test_pdf_zip_buffer.py (ZipExtFile-like object without chunks(); mocks fitz.open / TOC path)
• Maintainers: upload a PDF-only zip in the knowledge UI and confirm paragraphs appear; check logs if a bad member fails

AI disclosure

This pull request was prepared with assistance from an autonomous coding agent (Grok Bot). I reviewed the diff and take responsibility for the change.

[ump45nose]

Closing: opened against main by mistake. MaxKB default branch is v2, where PdfSplitHandle already uses hasattr(file, "chunks") + shutil.copyfileobj for zip members. Will re-check #7436 on v2 before opening a corrected PR.