Automunge

- ordinal processing for categorical data via 'ordl'
- numbercategoryheuristic repurposed for ordinal evaluation in evalcategory
- option to exclude ordinal columns from PCA transform with PCA_cmnd bool_ordl_PCAexcl
- "verbose" option via printstatus
- new method to process labels for validation sets, now incorporated into postmunge call
- Feature importance evaluation now uses validation set ratios passed to automunge
- a few improvements to labelfrequencylevelizer, especially for single column labels
- labelfrequencylevelizer works with ordinal labels

- fixed bug with 'null' processing (for deletion of set)
- removed all instances of SciKit's Labelencoder / OneHotEncoder functions, replaced with method making use of pandas
- set default feature importance evaluation to off since it requires inclusion of labels column in training set
- new processing functions for log transform and bins groupings by powers of 10,
processing categories log0, log1, pwrs

reverted one of the two edits from 2.20, much better now

Small corrections to the insertinfill function to address found bug associated with data type of inserted values.

Fixed an issue with passing non-contiguous integer indexes introduced in the 2.17 update.

New infill methods mean and median available for numeric sets.
User can now pass partial list of infill methods to automunge in assigninfill.
A fed clean ups in the code.