Update python-levenshtein to 0.21.1

[pyup-bot]

This PR updates python-levenshtein from 0.12.2 to 0.21.1.

Changelog

0.21.1

Changed
- upgrade ``rapidfuzz-cpp`` to ``v2.0.0``

0.21.0

Changed
- relax dependency requirement on ``rapidfuzz``

0.20.9

Fixed
- fix function signature of `get_requires_for_build_wheel`

0.20.8

Fixed
- type hints for `editops`/`opcoded`/`matching_blocks` did not allow any
hashable sequence

0.20.7

Fixed
- type hints did not get installed

0.20.6

Fixed
- fix incorrect result normalization in `setratio` and `seqratio`

0.20.5

Fixed
- fix support for cmake versions below 3.17
- fix version requirement for `rapidfuzz-cpp` when building against a previously installed version

0.20.4

Changed
- modernize cmake build to fix most conda-forge builds

0.20.3

Changed
- Added support for Python3.11

0.20.2

Fixed
- fix matching_blocks conversion for empty editops

Changed
- added in-tree build backend to install cmake and ninja only when it is not installed yet and only when wheels are available

0.20.1

Fixed
- fix broken matching_blocks conversion

0.20.0

Changed
- use `matching_blocks`/`apply`/`remove_subsequence`/`inverse` implementation from RapidFuzz

Fixed
- stop adding data to wheels
- fix segmentation fault on some invalid editop sequences in subtract_edit
- detect duplicated entries in editops validation

0.19.3

Added
- add musllinux wheels

0.19.2

Added
- add missing type hints

0.19.1

Added
- Add type hints

0.19.0

Changed
- implement all Python wrappers mostly with cython
- replace usage of deprecated Python APIs

Fixed
- fix behavior of median and median_improve

0.18.2

Changed
- Allow installation from system installed versions of `rapidfuzz-cpp`

0.18.1

Fixed
- Indel.normalized_similarity was broken in RapidFuzz v2.0.0 (see 20)

0.18.0

Fixed
* Fixed memory leak in error path of setratio

* Fixed out of bound reads due to uninitialized variable in median
* e.g. quickmedian(["test", "teste"], [0, 0]) caused out of bound reads

Changed
* Use a faster editops implementation provided by RapidFuzz
* Reduce code duplication
* reuse implementations from rapidfuzz-cpp
* Transition to scikit-build

0.17.0

* Removed support for Python 3.5

0.16.1

* Add support for RapidFuzz v1.9.*

0.16.0

* Add support for Python 3.10

0.15.0

* Update SequenceMatcher interface to support the autojunk parameter

0.14.0

* Drop Python 2 support
* Fixed free of non heap object due caused by zero offset on a heap object
* Fixed warnings about missing type conversions
* Fix segmentation fault in subtract_edit when incorrect input types are used
* Fixed unchecked memory allocations
* Implement distance/ratio/hamming/jaro/jaro_winkler
using rapidfuzz instead of providing a own implementation
* Implement Wrapper for inverse/editops/opcodes/matching_blocks/subtract_edit/apply_edit
using Cython to simplify support for new Python versions

0.13.0

* Maintainership passed to Max Bachmann
* use faster bitparallel implementations for distance and ratio
* avoid string copies in distance, ratio and hamming
* Fix usage of deprecated Unicode APIs in distance, ratio and hamming
* Fixed incorrect window size inside Jaro and Jaro-Winkler implementation
* Fixed incorrect exception messages
* Removed unused functions and compiler specific hacks
* Split the Python and C implementations to simplify building of
the C library
* Fixed multiple bugs which prevented the use as C library, since some functions
only got defined when compiling for Python
* Build and deliver python wheels for the library
* Fixed incorrect allocation size in lev_editops_matching_blocks and
lev_opcodes_matching_blocks

Links

• PyPI: https://pypi.org/project/python-levenshtein
• Changelog: https://pyup.io/changelogs/python-levenshtein/
• Repo: https://github.com/maxbachmann/python-Levenshtein