Remote Sensing Vision-Language Foundation Models without Annotations via Ground Remote Alignment