Float64sr round down for powers of two

[milankl]

julia> stochastic_round(Float64,Double64(2))
1.9999999999999998

because eps changes from [1,2) to [2,4) the eps calculated from 2 reaches into prevfloat(2.0) making it possible to round down from 2 at a chance of 1/4.

[milankl]

@avleenk2312 I realised this issue today, but you're not using Float64sr anyway right? Also I believe in practice it's of very little importance although it looks bad because the chance to hit exactly 2 as a result of an arithmetic operation in a simulation are usually quite low?