Code improvements

[krystian8207]

A list of simple todos extracted from code comments:

1. When using update_source (case when keep_na is integer or TRUE) use restore method based on limited state.
2. How can we allow to run add_step method when multiple filters are passed? Should we allow it?
3. Provide an option to save cohort state eveytime an operation changing state is run (e.g. add_step, rm_step etc and save states as a queue storing history).
4. In repro code generation we use multiple append calls. Maybe writing e.g. flatten_list method would simplify it?
5. When running .post_filtering we reassing the whole data. Can we optimize it to modify only "touched" datasets?
6. Source$update_filter - can we add more robustness when wrong filter ids are passed?
7. Can we extract choices from stats? Related to #47. When large dataset having only choices without computing stats can be faster.
8. Add column type - filter type validation (probably as optional as it needs some operations).
9. get_range_frequencies and get_date_range_frequencies - verify correctness and speedup (e.g. group_by can work slow).