Pygeoprocessing

------------------
* Fixing an issue with imports at the ``pygeoprocessing`` module level that was
causing linters like PyLint and IDE command-completion programs like JEDI-vim
to not be able to identify the attributes of the ``pygeoprocessing`` module
namespace.
* Reducing the amount of memory used by
``pygeoprocessing.new_raster_from_base`` when filling an array with values to
only as much memory as is needed for the datatype.
* Fixing a memory leak in ``pygeoprocessing.raster_calculator`` where
shared memory objects were being inadvertently created when they should not
have been and then they were not subsequently destroyed.
* ``calculate_disjoint_polygon_set`` will now skip over empty geometries.
Previously, the presence of empty geometries would cause an error to be
raised.
* Fixed a ``DeprecationWarning`` in ``calculate_disjoint_polygon_set`` caused
by the use of a deprecated logging ``warn`` method.

------------------------
* Post-release due to corrupted sdist released on Github and PyPI. The sdist
for 2.3.3 failed to install. The wheels were unaffected. No code changes.

------------------
* Fixing a bug in ``shapely_geometry_to_vector`` where a feature name mismatch
between the ``fields`` and ``attribute_list`` inputs would silently pass
under most circumstances. Now an informative ``ValueError`` is raised.
* Testing against Python 3.10.
* Pinned ``numpy`` versions in ``pyproject.toml`` to the lowest compatible
version for each supported python version. This prevents issues when
``pygeoprocessing`` is used in an environment with a lower numpy version
than it was built with (https://github.com/cython/cython/issues/4452).

------------------
* Restore functionality in ``reclassify_raster`` that allows for nodata
values to be reclassified. This was accidentally removed in 2.3.1.

------------------
* Slightly change the error message displayed for a
``ReclassificationMissingValuesError``.
* Handle float nodata values correctly in ``reclassify_raster``.
* Fix a bug in ``distance_to_channel_mfd`` where pixels that only partially
drained to a stream were given incorrect distances.
* Fixed an issue with ``raster_band_percentile`` that would raise an
exception if an input raster had an undefined nodata value and/or
would cause an invalid result if the raster contained non-finite
values.
* After many years with the Natural Capital Project, Rich Sharp has stepped
down from the Project and as the maintainer of ``pygeoprocessing``. James
Douglass is taking his place, and this change is now reflected in
``setup.py``.

------------------
* Added a ``single_outlet_tuple`` parameter to ``routing.fill_pits`` that
forces a DEM to only have one outlet at any point on the raster. The
fill effect is that all pixels will drain to the raster coordinate at
``single_outlet_tuple``.
* Added a ``detect_lowest_sink_and_drain`` function that finds the lowest
DEM pixel that drains to nodata/edge and the lowest DEM pixel that could
be a sink. The values that result from this call can be used to condition
a DEM that is known to have a single drain using the
``single_outlet_tuple`` parameter in ``routing.fill_pits``.
* Fixed a bug in ``routing.fill_pits`` that could cause the nodata region of
a DEM to be incorrectly filled with non-nodata values.
* Fixed an issue where ``create_raster_from_vector_extent`` would use an
approximation of the layer extent to infer the input vector extent. Now
uses the individual geometry to infer vector extents.
* Fixed a bug in ``routing.distance_to_channel_mfd`` where pixels that drained
off the edge of the raster had negative values rather than nodata values.
* Expanded the error message raised by ``transform_bounding_box`` when the
bounding box cannot be transformed to provide more helpful details.
* Add support and testing for GDAL 3.3.0.
* Fixed exception occuring in ``stitch_rasters`` when base rasters were
larger than the target raster on the ``y`` axis.