split_table: add support for multi-category stratification and other stratification options

[nbokulich]

Improvement Description
Stratify randomized splitting based on one or more sample metadata categories, or other options. Some example uses:

1. stratify by group (e.g., for cross-group or cross-batch prediction)
2. stratification of dependent samples (e.g., different sample types collected from same subject)
3. temporal stratification

Current Behavior
Stratification can only be done based on a single value, and in the pipelines (e.g., classify_samples) this is the target value.

Proposed Behavior
The stratify option should accept a list of column names, instead of a bool, possibly also additional options for more complicated stratification types (temporal, group kfold)

Comments
See sklearn stratification docs for some other examples....

References
forum x-ref