DQ-LoRe: Dual Queries with Low Rank Approximation Re-ranking for In-Context Learning