Automunge

- fixed outlier bug for PCA application involving inconsistent index numbers
- fixed bug which ommitted NArw columns (infill markers) from Binary dimensionality reduction
- added more detail to Binary dimensionality reduction printouts
- removed a superfluous copy operation that was possibly costing some memory overhead

- new dimensionality reduction technique, bulk transform of all boolean columns via binary encoding, available by passing Binary = True
- a tradeoff between positive aspects of memory efficiency / energy efficiency / number of weights vs perhaps some small impairment for ability of model to handle outliers, as now any single configuration of boolean sets not seen in training data will trigger an infill for the entire set
- I suspect this might take longer to train, as less redundancy in training data, but again energy efficiency in inference etc.
- fixed bug for overwriting default transform categories under automation

- transformation 'dxd2' is similar to 'dxdt', but instead of comparing singular cells accross a period time step, it compares the average of the sets of cells at 1 and 2 periods preceeding current time step
- such as to smooth / denoise data
- transformation category 'dxd2' family (dxd2/d2d2/d3d2) now accepts parameter 'periods' for number of time steps for evaluation
- such as may be useful for calculating velocity / acceleration / jerk over custom time steps
- useful for cumulative data streams

- transformation category 'dxdt' family now accepts parameter 'periods' for number of time steps for evaluation
- such as may be useful for calculating velocity / acceleration / jerk over custom time steps
- useful for cumulative data streams that may challenge current paradigms of deep learning

- Label Smoothing now supports application to multiple one-hot encoded sets originating from the same label source column
- (such as may be a product of our family tree primitives for applying sets of feature engineering transformations which may include generations and branches)

- Label Smoothing with LSfit now carries through fit properties to different segments of data (e.g. labels for train, test, or validaiton sets) for consistent processing based on properteis derived from distribution of label categories in the train set
- To carry through Label Smoothing parameters from train set to other segments pass LabelSmoothing parameters as True
- User now has ability to pass single column label sets to automunge(.) and postmunge(.) such as for processing labels independent of corresponding train sets
- Note accepts single column numpy arrays or pandas dataframes but (for now) not pandas series as input, so can just convert a series to dataframe with e.g. df = pd.DataFrame(df[column])
- fixed silly bug associated with passing empty assignparam to automunge(.)