Automunge

- a few small cleanups associated with validation sets
- removing code relics relevant to deprecated valpercent2

- small bug fix for inversion applied to subset of columns
- originating from parameter validation function
- for when passing postmunge inversion parameter as a list of columns

- important: backward compatibility breaking update
- reduced the number of sets returned from automunge(.) and postmunge(.)
- automunge returned sets reduced from 17 to 10
- postmunge returned sets reduced from 5 to 4
- function calls now look like:

train, train_ID, labels, \
val, val_ID, val_labels, \
test, test_ID, test_labels, \
postprocess_dict = \
am.automunge(df_train)


test, test_ID, test_labels, \
postreports_dict = \
am.postmunge(postprocess_dict, df_test)

- also deprecated automunge(.) parameters valpercent1, valpercent2
- replaced with / consolidated to valpercent
- so functions now only return a single validation set when elected
- (Had been so focused on retaining compatibility of tutorials published years ago lost sight of design principles. Came to my senses.)

- new auto ML option for ML infill by FLAML library
- available by setting ML_cmnd = {'autoML_type':'flaml'}
- parameters can be passed to fit operation by e.g.

ML_cmnd = {'autoML_type':'flaml', \
'flaml_classifier_fit' :{'time_budget' : 5}, \
'flaml_regressor_fit' :{'time_budget' : 5}}

- where time_budget specifies max seconds training each feature
- parameters default to verbose = 0
- FLAML library courtesy of Microsoft team
- Chi Wang, Qingyun Wu, Markus Weimer, Erkang Zhu, "FLAML: A Fast and Lightweight AutoML Library"

- small iteration associated with the new inplace parameter
- incorporated inplace into parameter validations
- also revised inplace to default to off when incorrect format passed

- new parameter inplace available for both automunge(.) and postmunge(.)
- defaults to False
- when True the df_train and df_test passed to automunge(.) or postmunge(.)
- are overwritten with the returned train and test sets
- inplace reduces memory overhead since don't have to redundantly hold data set in ram