Allow categorical transformers to ignore missing values for forward and reverse

[npatki]

Problem Description

For some types of usage, it is desirable for the transformers to ignore missing values for both the forward and reverse transform. For exmaple, maybe the downstream model or analysis actually is OK with missing values so we want to keep them.

Currently, all transformers have settings for missing value handling except for the categorical ones.

Expected behavior

The following applies to: UniformEncoder, OrderedUniformEncoder, LabelEncoder, OrderedLabelEncoder, and BinaryEncoder

These transformers should have a parameter called:

• missing_value_encoding: This parameter controls how missing values are encoded when transforming and reversing the data.
  • (default) "new_category": Missing values are encoded as a separate category. When reversing, the transformer reverses the encoding to regenerate the missing values. (Note that this is the status quo for all of these transformers.)
  • None: Missing values are not encoded. In this case, missing values are kept missing. When reversing, the transformer passes any missing values through as-is.

Note that for OrderedUniformEncoder and OrderedLableEncoder: It is ok if the order parameter does not include missing values if missing_value_encoding=None.