Ties should be approxmate in Esteemer

[pboisver]

We don't want artifacts like .499999999999998 < .500000000000002 to always force a particular result. We should have a sensible approach to determining if two scores are actually different.

This might mean that we think carefully about how we represent moderators when we detect a signal, but more likely we should think about what scale, range, and precision.

rounding may solve this (already).

[faridseifi]

We round final score to 1 decimal point and that takes care of ties.