More efficient way for split_x_c #100
Description
In the current mcpdft split_x_c implementation, the most time-consuming step, get_ontop_pair_density, will always be performed twice. When the x and c part shares the same Pi (say tGGA_x+tGGA_c, or ftGGA_x+ftGGA_c, etc.), the procedure can probably be optimized to "get_ontop_pair_density once, then eval_ot twice", which saves roughly 50% time.