This FPE was first seen in r22 (https://its.cern.ch/jira/browse/ATR-23551), but is present in both r21 and r22. This commit is the r21 fix, which can then be cherry-picked into r22. Note that the choice was made to keep the same behaviour as the current tool, which looked for NaNs instead of preventing their creation in the first place. That is, negative logarithms result in a calibration factor of 1 (no calibration) or a calibration weight of 0 (for combined mass calibrations).