Deduplicate hybrid search results in LanceDB

Resolved an issue where LanceDB's search() could return the same document multiple times when hybrid search retrieved it via both vector similarity and full-text search. Results are now deduplicated before being returned, ensuring each document appears only once regardless of which search path surfaced it.

‍

Details:

Fixes duplicate results in hybrid search caused by the same document matching both the vector and FTS indices
Deduplication is applied automatically; no configuration changes required
Improves result quality and reduces noise for agents and workflows using LanceDB hybrid search