* rename get_size to axes_rank and move axes buffer to axes.hpp * added fill experiments * faster and simpler fill_impl * faster specializations for linearize_value