test: regression coverage for connected image/table hydration

[Frankie-Xu]

Summary

• Test-only: add synthetic regression coverage for connected image/table hydration in assemble_retrieval_results.
• No production behavior change, no API / worker / deploy / migration impact.
• Related to Connected image chunks silently dropped in assemble_retrieval_results #206 (maintainers can decide whether to close).

The silent drop of connect_to image targets was fixed on main by #221 (cb807d18). These tests lock:

• text → image is inlined as [Image: {asset_url|file_path}] plus description
• text → table still composes summary/keywords/caption (HTML not leaked)
• connected targets are excluded from the standalone result list
• public source projection keeps page_nums on page rows and related asset fields (file_path, source_chunk_path, inlined image URL) on the parent

Verification

Red/green check (temporarily restored the old if target_type != 'table': continue gate in _connected_media_parts, then reverted with git checkout):

• red: 4 failed, 1 passed — all image assertions failed; table composition still passed
• green (restored): 5 passed in packages/shared-python/shared/tests/test_connected_image_hydration.py

Commands:

make lint                                          # All checks passed
make typecheck                                     # 0 errors, 0 warnings
uv run pytest packages/shared-python/shared/tests/test_connected_image_hydration.py -q
                                                   # 5 passed
uv run pytest packages/shared-python/shared/tests -q
                                                   # 62 passed
uv run pytest apps/worker/tests/contract -q        # 303 passed

Not tested: live documents, LLM, storage, or Postgres hydration of missing connect_to targets (db=None in these unit tests).

Deployment Notes

• No new or changed environment variables.
• No database migrations, queue, storage, or release-order requirements.
• Backwards compatible (tests only). Rollback is revert of the test file.

Checklist

• Tests were added or updated when behavior changed
• Public docs, examples, or OpenAPI contracts were updated when needed (N/A — test-only)
• Database migrations are idempotent and safe to deploy (N/A)
• Logs, errors, and validation paths avoid leaking secrets or user data
• The pull request description explains any breaking or user-visible change (none)

Made with Cursor