Chebyshv filtering and k-point parallelization

The chebyshev filtering seems to have a dependence on the k-point parallelization. This seems to be due to the generation of the random vector used to estimate the upper bound.

I am implementing a fix for this.