[WeakMultiLabels] Merge with WeakLabels using a 2 dim wl matrix

[dcfidalgo]

The current implementation of the weak label matrix for multi-label classifications is overcomplicated (3-dim), and the MajorityVoter does not take advantage of the format. I propose to simplify this, basically treating rules with x labels as x rules, and remove the WeakMultiLabels class.

Pros: reduce code complexity, get the extend_matrix functionality also for multi-label cases.
Cons: if at some point we find a label model that can take advantage of the 3-dim format, we must implement a method that transforms the 2-dim format.

[dcfidalgo]

I was looking into this in more detail, and I decided to keep the WeakMultiLabel class for now. It's harder than I thought to move it to a 2dim matrix, and I'm not sure anymore if it actually would reduce the complexity.

It was, however, relatively easy to bring the extend_matrix functionality to the 3d weak label matrix, see #1577 .

I will close this issue for now.