Bug fix: filter out nans in npc combining functions

[sequoiarose]

Currently the fisher and other combining functions do not filter out nan pvalues, which results in the function returning nan if a single p value is nan

[pbstark]

P-values should be in [0, 1] and never NaN, so I think the default behavior of combining functions should be to throw an error if there are NaNs, with a flag to either ignore silently or ignore and warn if there are NaNs among the P-values.