Handle sets of NaN fluxes properly in coaddSpectra.py

I see coaddSpectra.py complaining:

WARN: Multiple i2_hsc flux for target catId=10011 tract=1 patch=1,1 objId=75338845972883613 ({nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan, nan}); using average

and this appears to be coming from the (correct) IEEE754 behaviour of e.g. np.float64 that NaN != NaN. However, in this context, it's unhelpful.

Please fix this so that all {{NaN}}s test equal, and we avoid this duplication.