Document multi-vector embeddings and late-interaction search

[lukekim]

Summary

Documents multi-vector (column-of-vectors) embeddings and ColBERT-style late-interaction search — features that exist in spiceai but were previously undocumented.

Changes

• New page website/docs/features/search/multi-vector.md — multi-vector vs chunking, configuration, aggregation strategies (max/mean/sum), max_elements_per_row, SQL usage, ColBERT-style late-interaction multi-query search, passthrough schemas, and limitations.
• Search index (features/search/index.md) — added Multi-Vector as a first-class search method with an example.
• Vector search (features/search/vector-search.md) — cross-reference to the new page.
• Embeddings component (components/embeddings/index.md) — new Multi-Vector Embeddings section covering list-typed source columns and the aggregation / max_elements_per_row knobs.
• Datasets reference (reference/spicepod/datasets.md) — documented the previously missing columns[*].embeddings[*].aggregation and columns[*].embeddings[*].max_elements_per_row fields.
• SQL search reference (reference/sql/search.md) — documented the multi-query (late-interaction) form of vector_search.

[github-actions]

✅ Pull with Spice Passed

Passing checks:

• ✅ Title meets minimum length requirement (10 characters)
• ✅ Has at least one of the required labels: area/blog, area/docs, area/cookbook, dependencies
• ✅ No banned labels detected
• ✅ Has at least one assignee: lukekim

[github-actions]

🚀 deployed to https://ca151686.spiceai-org-website.pages.dev