Ethicml

A couple of touch-ups. Updated dependencies; Made seed a parameter to all models; The docs now auto build on update.

We now have the option to just refer to the `S` variable in the produced reulsts as just `S` instead of this changing per dataset. This is useful if you're looking at runs across multiple datasets and want the produced .csv to be easier to navigate.

Minor update of dependencies and tabular datasets now also accepting enums.

Added the LSAC LAw Admissions Dataset form Counterfactual Fairness by Kusner et al.

it's about time we went to 0.3.
This is mostly just updates to dependencies and getting some parity across datasets (all datasets now have discrete feature groups in case you want more information about how a decoder should reconstruct them).

Also, this release was made with our lovely new release script - thanks thomkeh 🙏 , so releasing smaller updates more often should be easier.

Changes

- Add `labels_as_featurse` option to `dataset.load()`
- Expose constants that determine where the celeba dataset is stored