Function execute_query 

pub async fn execute_query(
    csv_path: &str,
    sql: &str,
) -> Result<QueryResult, ComputeError>

Execute SQL against a CSV path (optionally via a Parquet cache) and return a rich result.

Optimizations:

• CSV > 100 MiB → convert/reuse sibling .parquet when present
• SELECT * without LIMIT → append LIMIT 1000
• Results are streamed via DataFusion execute_stream to limit RAM spikes