Fix the usage of r_w across the crate.

[CPerezz]

case A) In case that we use randomness for the commitments, r_w should be not lost, as later is needed to be folded together when prover folds the witness.
.>
As an example, in multifolding-poc it was returned together with the witness at the end of the method to define the LCCCS:
https://github.com/privacy-scaling-explorations/multifolding-poc/tree/d118ea56418dadde005b5854b5cbbf38a686d24d/src/ccs/lcccs.rs#L66

case B) In case that we don't use randomness for the commitments, we don't need r_w at all and can just be removed. (which, seeing the pedersen::CommitmentEngine, it does not use any randomness internally for the commitment.

So, if we go with 'case A', we should return r_w at the end of the method (together with the witness) and probably update the Pedersen implementation to support using randomness, and if we go with 'case B' we don't need to define _r_w.

Originally posted by @arnaucube in #41 (comment)

As @arnaucube correctly stated above, we should take a decision and either remove completelly the usage of randomness for the Pedersen Commitments, or, change the current impl of the Pedersen Commitments and actually include the randomness support for them.