Runs SQL against a Databricks SQL warehouse via the Statement Execution API, streams the Arrow-IPC result chunks with backpressure into DuckDB for JSON/rows/Arrow serialization -- no pyarrow required.