Making the epsilon for the sampler increase and better Z generation

In this merge request I am tackling mostly 2 problems:

• During the creation of the condition for the sampler, there is an epsilon that allows for some range, if not it would be impossible to obtain samples in the range. If the requested values are too challenging, this conditions might be more difficult to meet. In this case, the epsilon value is increased during some iterations, until the conditions are met.
• Now, the sampling procedures include a breaking condition, in case some maximum iterations are reached.
• The generation of Z needs to be conditioned on the full vector Y, not only the requested values. Now, after having the samples for Y, we generate a Z condition on each of this Y. And this is then fed to the decoder for generation.

We still need to assess more exhaustively how much the results improve in generation, in terms of error.

Besides, some other small implementations, such as the harmonization of the names for the summaries.