Would it make sense to have 1±NaN > 0 return False±Nan?

[lebigot]

trafficstars

In order to catch more violations of the linear approximation hypothesis that underlies the calculations in the uncertainties module, it would make sense to have 1±NaN > 0 return False±Nan.

Now, this mixes booleans and floats, so this looks a little strange (even though booleans are, in Python, integers). Print formatting routines would probably have to be extended so as to handle this case.