feat(embed): Implements multimodal image-vector retrieval

[chengjoey]

Implements multimodal image-vector retrieval

• Add a MultimodalEmbedder interface and a multimodalEmbedder decorator (chat / sglang envelopes, one embedding request per image). Text calls still go through the catalog protocol, so only the multimodal call shape changes.
• New chunk type image_vector plus an indexing_strategy.image_vector_enabled switch. It is decoupled from VLM: images embedded inside docs (PDF/Word/Markdown) are indexed without turning on image processing.
• Indexing and query sides share a single usesMultimodalEnvelope predicate. A text query is encoded through the same envelope and recalled via ANN in the same collection; image_vector chunks are exempted from reranking (their content is only a markdown image link).
• Flipping the switch on an existing KB enqueues kb:reindex_vectors to backfill image vectors.

Bug fix

writeImageVectorChunk omitted IsEnabled when constructing IndexInfo, so the image vectors were persisted with is_enabled = false and silently filtered out by every retrieval engine — written but never recallable. Fixed and covered by a regression assertion (with mutation verification).

Verification

hybrid-search with query_text="图片里有什么" successfully recalled all 4 images cross-modally.

Description

Type of Change

• 🐛 Bug fix
• ✨ New feature
• 💥 Breaking change
• 📚 Documentation update
• 🎨 Refactor
• ⚡ Performance improvement
• 🧪 Test
• 🔧 Configuration / Build / CI

Related Issue

feat: #2875

Checklist

• git diff --check origin/main...HEAD passes
• Changed source files are formatted
• Targeted tests for the changed packages/components pass
• Diff-scoped lint passes where applicable (for Go: golangci-lint run --new-from-rev=origin/main ./...)
• Full-repository checks were run, or any unrelated/environment-dependent failures are documented above
• Self-reviewed the code
• Added/updated tests covering the change
• Updated related documentation (README, website-docs/, Swagger annotations, etc.)
• Breaking changes are clearly called out in the description above

Screenshots / Recordings