Smarter Query Generation with the Semantic Metadata Layer (SML)

**Value proposition**
Agents today spend tokens at query time exploring index mappings to understand data structure. Agent Builder is introducing a "Semantic Metadata Layer" which pre-catalogs this knowledge, making agents faster, cheaper, and more accurate when querying Elasticsearch data.

**Expected outcome**
Agents automatically discover available indices and data streams, extract semantic metadata (field meanings, data types, sample values), and use this to write accurate ES|QL queries without runtime exploration. Users can reference specific data sources via `@` mentions in chat. ES|QL generation is also optimized for speed and token efficiency, including a "fast mode" for single-pass generation.

**Key user stories**
- As a data analyst, I want the agent to automatically know the structure of my sales data index so it writes correct queries the first time.
- As a data analyst, I want the agent to learn how I query data to improve the quality and efficiency of my results with subsequent interactions
- As an operator, I want ES|QL generation to be faster and cheaper without sacrificing accuracy.