minor issue: acc_hold cohort variables should not be zero'd

[rgknox]

Variables labeled "_acc_hold" are being zero'd in the group zero'ing calls. These variables are not built or iterated, so they don't need to be zero'd, its safer to leave them as nan's so they aren't used without proper initialization. They were indeed being used in the history routine, and in the routine where we calculate a smoothed NPP (for fixation calculations in the hlm), and newly created cohorts that had not experienced a day of productivity were being asked for these values (inappropriately). This will slightly under-bias the history output values since the zeros will be mixed in.