- remove `class PriorityQueue`
- use `prioritized_metadata` DataFrame in place of `queues_per_group`
- repurpose `create_queues_per_group` to `create_sizes_per_group`
- other logical refactoring:
    - use global dummy group key and value
        - key is `list`: pd.DataFrame.groupby does not take a tuple as grouping key, also our `--group-by` is stored as list already.
        - value is `tuple: `get_groups_for_subsampling` currently returns group values in this form.
    - use records_per_group for _dummy
        - replace conditional logic of `records_per_group is not None` with `group_by`
- add functional tests

…g logic

This test currently fails with a pandas-specific index error.

Instead of calculating a new (year, month) tuple when users group by
month, add a "year" key to the list of group fields. This fixes a pandas
indexing bug where calling `nlargest` on a SeriesGroupBy object that has
a year and month tuple key for month causes pandas to think the single
month key is a MultiIndex that should be a list. Although this commit is
motivated to fix this pandas issue, this implementation of the
year/month disambiguation is simpler and a more idiomatic pandas
implementation that wouldn't have been possible in the original augur
filter implementation (before we switched to pandas for metadata
parsing).

Simplifies unit tests and doctests by expecting a single value for each
month instead of a tuple.