Unification of handling weights passed as arguments to functions

[msluszniak]

As mentioned here we want to unify the way we pass the weights in Scholar to the algorithms. Currently, there are a few ways that we do it. For example, we pass them as tensors or as a list via opts.

[JoaquinIglesiasTurina]

I would like to take a stab at this one, particularly regarding linear methods.
I think this would also be a great opportunity to DRY up some of the duplication of private functions that's spreading over there.
So far, the better approach I've found is to move all these functions to a helper module, make them public,
but make sure to set @moduledoc false and @doc false. My proof of concept is here.
Is this the best approach to handle this? Or should I take an alternative approach?

[msluszniak]

Yes, this looks great! You can send a PR with your improvements :)