Set epsilon to zero in `RobustMAD`

[niranjchandrasekaran]

@gwaygenomics You may already know that we sometimes have had issues(private repo) with mad_robustize where one or few features end up being significantly larger than the other features, and then it affects the downstream metrics that we compute. @shntnu said this behavior can be overcome by setting the epsilon variable to 0 instead of 1e-18 that it is set to, by default. Currently, epsilon cannot be set while calling normalize() but we could do one of the following

1. Change the default value of epsilon. This is the easiest option, but it is possible that it will break something.
2. Allow setting epsilon while calling normalize().

Which one would you suggest?