Data-adaptive Differentially Private Prompt Synthesis for In-Context Learning