Learning rate matters: Reexamining optimal power expansion planning with endogenized technological experience curves