Enforce the bounds more accurately on the relational implementation

[salmans]

Currently the bounding a model in the relational implementation has two issues:

• It doesn't apply the bound when adding every new observation: it keeps augmenting the new model fully at every chase-step.
• It doesn't take the equivalence classes of elements into account when calculating the "size" of the model.

Idea:

• keep track of model rewrites by references to elements (in Rc<RefCell<E>>).
• compute the size of models based on the values of rewrites.
• check the bound before adding every new observation (add the observation to the model if that particular observation doesn't violate the bound)

Originally posted by @salmans in #104 (comment)