Add zip/archive recursive ingestion for KB document uploads #16774

Problem

AutoBot has no path anywhere that opens a .zip archive and ingests its members as documents. Confirmed absent: grep -rli "zipfile" across autobot-backend/media/ and autobot-backend/knowledge/ returns no hits; ALLOWED_EXTENSIONS in autobot-backend/api/knowledge.py:126 has no .zip entry; DocumentExtractor.SUPPORTED_FORMATS (autobot-backend/utils/document_extractors.py) has no zip entry either. A user with a folder of documents must upload each file individually.

Location

• autobot-backend/api/knowledge.py:126 (ALLOWED_EXTENSIONS)
• autobot-backend/utils/document_extractors.py (DocumentExtractor.SUPPORTED_FORMATS)
• New extractor, likely autobot-backend/media/document/archive.py, following the existing module split in media/document/ (extraction.py, provenance.py, ocr.py)

Impact

Bulk KB uploads (a folder of research PDFs, a project's docs) require N individual uploads today instead of one zip upload — a real workflow friction for any user onboarding an existing document set.

Acceptance Criteria

• A .zip upload is unpacked and each supported member is routed through the existing per-format extractors (extract_document, DocumentParser) — no new parsing logic duplicated
• Hard caps enforced and configurable via the existing env-var-backed-constant pattern (see min_text_page_ratio()/max_table_pages() in media/document/extraction.py for the pattern to follow): member count, nesting depth, and total decompressed size — this is a decompression-bomb vector and ships with the caps, not as a follow-up
• Each member's extraction respects the existing per-file extraction_timeout() (media/document/ocr.py) individually, not one timeout for the whole archive
• An unsupported or corrupt member inside the archive is skipped and reported, not aborting the whole batch (matches existing per-file failure handling in DocumentExtractor.extract_from_directory)
• Test coverage includes a zip bomb / oversized-member fixture asserting the caps actually trip

Discovered During

Research pass — docs/research/document-to-markdown-conversion-pipeline.md, "What We Can Adopt" #2