Dataiter

=========================

* `DataFrame.count`: New method, shorthand for
`data.group_by(...).aggregate(n=di.count())`
* `Vector.rank`: Handle empty and all-NA vectors

=========================

* `USE_NUMBA_CACHE`: New option, read from environment variable
`DATAITER_USE_NUMBA_CACHE` if exists, defauls to `True`
* Fix a possible issue with Numba caching

=========================

* Use `numba.extending.overload` instead of the deprecated
`numba.generated_jit`

=========================

* `DataFrame`: Don't try to do joins on NA values in `by` columns
* `DataFrame.drop_na`: New method

=========================

* `DataFrame`: Truncate multiline strings when printing
* `DataFrame.from_arrow`: New method
* `DataFrame.read_parquet`: New method
* `DataFrame.to_arrow`: New method
* `DataFrame.write_parquet`: New method
* `read_parquet`: New function
* `Vector.__init__`: Fix type guessing when mixing Python and NumPy
floats or integers and missing values
* Allow using a thousand separator when printing numbers,
off by default, can be set with `dataiter.PRINT_THOUSAND_SEPARATOR`

=========================

* Fix printing really small numbers