ATLASPHT15 fix
Created by: wilsonmr
I decided to make a separate PR for each DS, considering we want to make sure there are no issues this time.
The test passes now with this dataset and, as expected, the result is pretty much unchanged: https://vp.nnpdf.science/p6hYntpiRr2LB_89SjUFjw==
compared to the chi2 using the old dataset: https://vp.nnpdf.science/5Vo-7kp5QymErUprbZkWxw==/#table