UGFF-VLM: Uncertainty-Guided and Frequency-Fused Vision-Language Model for Remote Sensing Farmland Segmentation | Synapse