Extend the CRUD service by integrating a Chroma vector store to support semantic search over documents. Tasks: a) define the collection schema and how embeddings are produced; b) implement endpoints to upsert, delete, and query by vector similarity; c) implement a search_query function that returns a specified response object with a 'results' list that may be empty when there are no matches (never return None), and include fields like scores and metadata; d) describe how you would handle indexing, pagination, filtering by metadata, and eventual consistency; e) discuss latency targets, batching, and error handling in production.