Datgan

* Rewrote the mini batch discriminator to allow for larger batch sizes.

* Changed noise to take into account the number of categories in the categorical columns.
* Added noise to categorical conditional inputs.
* Removed the direct ancestors from the attention vector.

* Added the option to sample data from the DATGAN without randomizing the conditional inputs.

* Fixed an issue when transforming the DAG according to the conditional inputs.

* Add possibility to ignore some columns while assessing the results
* Fixed an issue with the DAG when using conditional inputs
* It was removing the edges between variables in the conditional inputs, thus leading to worse results since
some of the variables in the conditional inputs were not taken into account in the attention vectors.
* Revert to using the direct output of the LSTM cells for the attention vectors.

* Updated the README based on comments given by users.