Add tests for the vacuum-failure warning, documents deleted mid-rebuild,
chunkless documents, batch flushes in the embed-only, descriptions and full
paths, a missing docling blob under rechunk, missing and recoverable picture
bytes, and the source-missing fallback. Direct-call tests cover the staging
helpers and the idempotent phase-1 marker.
processing.chunk() merges text chunks with one synthetic Chunk per PictureItem-with-bytes,
sorted by iterate_items() position so chunk.order is structural.
embed_chunks dispatches on a Chunk._picture_data PrivateAttr
(text through embed_documents, picture through embed_image_query)
When a database was created with one embedding model and rebuild
--embed-only was run with a different model, it failed with a
vector dimension validation error.
- Store reads stored vector_dim when opening existing databases
- _rebuild_embed_only recreates chunks table to handle dimension changes
- Add test for rebuild with changed vector dimensions
