Phenonaut

Added
- orient argument to write_boxplot_to_file allowing horizontal or vertical plotting.

Added
- Ability to supply features argument to new Datasets rather than wrapping in metadata dictionary
- data property to Phenonaut objects to return phe[-1].df[phe[-1].features].values
- weights argument to KNNImputer

Changed
- Percentile argument for percent_replicating and percent_compact is no longer required, and inferred to be 95th (if higher is better), or 5th if lower is better

Fixed
- Broken test caused by tempfile creation and phe.revert

Added
- Added imputers (SimpleImputer, KNNImputer, RFImputer) to transforms.imputers

Added
- get_dataset_names to phenonaut.Phenonaut objects

Fixed
- PackagedDataset loaders do not keep a H5store open, allowing multiple concurrent runs in HPC environment.

Changed
- percentile_cutoff now optional for percent_replicating and percent_compact, allowing inference if None for 95th/5th percentile depending on similarity metric (higher or lower better)

Added
- PackagedDataset for BROAD-MOA assignment, derived from the LINCS Cell Painting project

Added
- subtract_func_results_on_features to phenonaut.data.Dataset, which allows subtraction of func results to samples (reworked subtract_median and subtract_mean to call this generic function)

Fixed
- Percent replicating docstring