Important fix to single top data implementation

This PR will (most likely) remove the build failures observed in #648 (closed) . It removes a bug in the commondata files for the single top differential distributions at 7 TeV from ATLAS, where the additive version of the systematic uncertainties (which is never used in the default systypes) were incorrectly computed. A similar fix in buildmaster NNPDF/buildmaster#131 has already been merged. I don't have enough words to apologise for this.