Update IC calculation to use "pseudocounts"

[balhoff]

Currently a term without instances is given the max IC of its immediate parents, so that we don't need to divide by a 0 frequency. Instead, @hlapp suggests we can give all terms a freebie +1 to their annotation count. This will also provide a slight bump to IC of child nodes when their parent has the same frequency.