Regression from "Fixed unsafe floating point comparisons"

The following discussion from !1490 (merged) should be addressed:

• @clemenci started a discussion:

  This change introduced a regression because in most cases the range is not actually defined for an NTuple so the condition is usually false and the branch effectively taken is

        lowerRange = 0;
        upperRange = -1;

  Also, in case of a double column upperRange = it->range().upper(); resolves to the equivalent of long upperRange = std::numeric_limits<double>::max(); which leads to an FPE.