Automunge

- new convention for user passed transformdict
- now user is able to pass partially populated family trees
- and any missing primitives are automatically added internally
- just requires a minimum of one populated primitive for each root category
- this makes user specification much cleaner / less typing
- also added a new validation for format of transformdict
- ensuring valid primitive spelling, data types, etc

- a small tweak to last rollout
- further testing demonstrated that randomseed range limited to 0:2**32-1
- (we had implemented range of 0:10**12)
- causing a bug when seed sampled in upper range
- found and fixed
- also moved the randomseed initialization to a slightly more approrpiate location in flow

- new default for automunge(.) randomseed parameter
- now set as False signalling application of a random random seed
- otherwise can still pass as an integer for specific desired random seed

- fixed imports for AutoGluon ML infill option
- as it appears they recently revised some of their imports for tabular

- a small bug in assignnan missing data injections found and fixed
- (realized injections were being redundantly applied instead of once per target)
- also found a poorly executed edge case disparity between automunge and postmunge
- associated with null category
- which is for columns that are simply deleted
- such as is default for training data sets containing all nan values
- so conformed treatment to be consistent with other transformation categories
- now null columns are deleted in circle of life function instead of special treatment
- also updated the column_map report returned in postprocess_dict
- so that it now contains entries for source columns that did not return any sets
- also small tweak to align flow for master infill functions for automunge and postmunge

- for assignnan missing data injections, we were basing on a random seed
- which was causing concurrent appplications to duplicate row configurations
- which was unintentional
- so struck the random seedings from these injections
- resulting in a random random seed between columns
- (missing data injections are supporting some experiments and demonstrations)
- also found the onh2 root category was redundant with onht
- so struck onh2 from library and read me