Use positive deviations and not absolute deviations to flag CR hits

[mgennaro]

Currently the CR hits flagging algorithm looks for values of the absolute deviations of the residual w.r.t. the estimated true flux.
This might lead to flagging negative deviations as CR which is unrealistic