Better define `cell_weights` in mesh-based regularizations

[santisoler]

What the cell_weights argument in mesh-based regularizations should be? We should better define them.

Because of their name, I think they should be related to the mesh cells and not necessarily to the model. So they could be an array of the size of the active (or every) cells in the mesh.

But that would leave us the question of how to set weights for the model values and not necessarily for the (active) cells.

Just leaving some thoughts here.