Jsonpickle

======
* **Breaking Change**: Python 3.7 is no longer supported.
* **Breaking Change**: Support for pre-0.7.0 ``repr``-serialized objects is no
longer enabled by default. The ``safe`` option to ``decode()`` was changed from
``False`` to ``True``. Users can still pass ``safe=False`` to ``decode()`` in order
to enable this feature for the purposes of loading older files, but beware that
this feature relies on unsafe behavior through its use of ``eval()``. Users are
encouraged to re-pickle old data in order to migrate away from the the unsafe loading
feature. (+514)
* The pickler no longer produces ``py/repr`` tags when pickling modules.
``py/mod`` is used instead, as it is clearer and uses one less byte. (+514)
* The test suite no longer uses the deprecated ``datetime.datetime.utcnow()``
function. (+539)

======
* The breaking changes from v4 were inadvertedly included in v3.4.1, which has
been yanked. This release remedies this by reverting the v4 changes.

======
* Support decoding pandas dataframes encoded with versions 3.3.0 and older. (+536)

======
* Officially support Python 3.12 in the GitHub Actions testing matrix, and update
GHA package versions used. (+524)
* Improve reproducibility of benchmarking commands on Linux by using taskset and
adding a "HOWTO" run benchmarks section in ``benchmarking/README.md``. (+526)
* The ``setup.cfg`` packaging configuration has been replaced by
``pyproject.toml``. (+527)
* ``yaml`` is now supported as a jsonpickle backend. (+528)
* `OSSFuzz <https://github.com/google/oss-fuzz>`_ scripts are now available in
the ``fuzzing/`` directory. (+525)
* Pure-python dtypes are now preserved across ``encode()``/``decode()`` roundtrips
for the pandas extension. (407) (+534)
* Pandas dataframe columns with an ``object`` dtype that contain multiple different
types within (e.g. a column of type ``list[Union[str, int]]``) now preserve the types
upon being roundtripped. (457) (358) (+534)
* Fix warnings in the test suite regarding numpy.compat usage. (533) (+535)

======
* The unpickler was updated to avoid using ``eval``, which helps improve its
security. Users can still pass ``safe=False`` to ``decode`` to use the old
behavior, though this is not recommended. (+513)
* Objects can now exclude specific attributes from pickling by providing a
``_jsonpickle_exclude`` class or instance attribute. This attribute should contain
the list of attribute names to exclude when pickling the object.

======
* A bug with the incorrect (de)serialization of NoneType objects has been fixed.
(+507)
* ``tests/benchmark.py`` was updated to avoid Python 2 syntax. (+508)
* The unpickler was updated to avoid creating temporary functions. (+508)
* Some basic scripts have been made to analyze benchmark results. (+511)
* Fix test suite compatibility with Numpy 2.x (+512)
* `setup.cfg` was updated to use `license_files` instead of `license_file`.