Ranx

Added
- `Run` now has an additional property to store metrics standard deviation.
- `evaluate` now has `return_std` flag to compute metrics standard deviation.

Changed
- `Qrels.from_df` now checks that scores are `numpy.int64` to avoid errors on Windows.
- `Run.from_df` now checks that scores are `numpy.float64` to avoid errors on Windows.

Changed
- All `Run` import methods allow for specifying the `name` of the run.

Fixed
- Fixed misleading error messages when importing `Qrels` and `Run` from `pandas.DataFrame` with wrong `dtypes`.

Added
- Added support for importing qrels from `parquet` files in `qrels.py`.
- Added support for importing runs from `parquet` files in `run.py`.
- Added support for exporting qrels as `pandas.DataFrame` in `qrels.py`.
- Added support for exporting runs as `pandas.DataFrame` in `run.py`.
- Added support for saving qrels as `parquet` files in `qrels.py`.
- Added support for saving runs as `parquet` files in `run.py`.

Fixed
- Fixed `f1` when there are no relevants.

Changed
- Moved `numba` threading layer settings to `ranx/__init__.py`.

Removed
- Removed dependency from `pytrec_eval`.

Added
- Added support for gzipped TREC files to `from_file` in `qrels.py`.
- Added support for gzipped TREC files to `from_file` in `run.py`.
- Added `name` parameter to `from_file` in `run.py`.

Fixed
- Fixed `rank_biased_precision` considering relevance as binary instead of graded.
- Fixed high memory consumption for `qrels` and `run`.

Fixed
- Fixed missing metric labels for `dcg` and `dcg_burges` in `report.py`.