Improve bin handling for discrete parameter types

[benkrikler]

Imported from gitlab issue 20

Currently there is no way to pre-inform carpenter what values to expect for discrete variables, eg. categoricals, natural numbers etc. This results in bins that would be expected, not being present in the resulting dataframe, especially for natural number variables. The config file specification should be expanded to handle these situations better.

First raised within CMS-CHIP: https://gitlab.cern.ch/cms-chip/chip/issues/7

[benkrikler]

So two new options seem sensible to me:

1. ranges for ordered but discrete parameters (e.g. number of hits in an event, but clipped to between 0 and 20)
2. a list of expected values (e.g. category names)

For 1:

• https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.clip.html#pandas.Series.clip
• Above method seems to set values out of the clipping range to the upper / lower bound -- is this what's wanted? Maybe set to nan instead, but then needs another approach.

For 2:

• https://stackoverflow.com/questions/19960077/how-to-implement-in-and-not-in-for-pandas-dataframe
• So need to use either isin or query which builds a mask for values then set all other values to nan or otherwsie.