Fix pretty subtle "bug" #110 in fmod

[pavpanchekha]

Basically, the fmod code sometimes involves calling fabs and it also sometimes depends on the precision of its arguments for intermediate computations. But it turns out that fabs sometimes returns 0 (for intervals that cross 0) and this 0 is the wrong precision (too high), which breaks refinement. The solution is to make a 0 of the correct precision.