Advancing Volumetric Medical Image Segmentation via Global-Local Masked Autoencoder