Phenonaut

Added
- Added filter_on_identifiers method to phenonaut.data.Datasets, allowing easy filtering of train/val/test sets
- Added filter_datasets_on_identifiers method to phenonaut objects to filter all datasets
- Added ContrastiveEncoder
- Added the shrink method to phenonaut objects and datasets
- Added calculation of type2 errors for permutation testing distinctness
- Added get_unique_treatments, and num_features properties of datasets
- Added ability to iterate Phenonaut objects, returning their datasets
- Added ability to call len on Phenonaut objects, returning the number of datasets they contain

Changed
- Refactored phenonaut.metrics
- Applied ruff checks/fixes

Fixed
- Exclude pptx files from black formatting
- Dataset constructor raises an error if features passed as metadata are not of type list
- merged dataset deletion with passed datasets
- test using dataset_groupby, corrected to groupby_datasets

Fixed
- RandomForest regressor no longer uses auto max_features hyperparameter, making it compatible with scikit-learn 1.1 onwards.

Changed
- The way versioning works internally to Phenonaut.
- Updated pyproject.toml to include powerpoint templates

Added
- New class of Error NotEnoughRowsError added to better flag runtime errors.
- Added checks to mp_value_score which ensure grouped dataframe groups have at least 3 rows required for calculations.
- In mp_value_score, groups with < 3 rows may be ignored by calling the function with raise_error_for_low_count_groups = False, in which case np.nan values will be returned for the group.

Changed
- Improved tests for mp_value_score, checking for correct behaviour within small groups with <3 rows.

Changed
- Refactored package to use pyproject.toml for install/build
- Updated classifier hyperparameters, `max_features='auto'` has been deprecated in scikit 1.1 and will be removed in 1.3. Now explicitly set to `max_features='sqrt'`.
- Removed progressbar2 in favour of tqdm throughout.

Fixed
- predict.profile updated to work with newer pandas
- Phenonaut.merge_datasets now honours the remove_merged flag