EXPO-FT: Sample-Efficient Reinforcement Learning Finetuning for Vision-Language-Action Models