Pixel-by-Pixel Cross-Domain Alignment for Few-Shot Semantic Segmentation