Taskgraph

-------------------
* Adding ``pyproject.toml`` for our build definitions.
* Python 3.6 has reached end-of-life and is no longer maintained, so it has
been removed from the automated tests.
* Python 3.7 has reached end-of-life and is no longer maintained, so it has
been removed from automated tests.
* Python 3.11 has been released, so ``taskgraph`` is now tested against this
new version of the language.
* Python 3.12 has been released, so ``taskgraph`` is now tested against this
new version of the language.

-------------------
* Testing against python 3.10 in github actions and officially noting support
for 3.10 in ``setup.py``.
* Testing against python 3.9 in github actions and noting support in
``setup.py``.
* Fixed an issue where exceptions raised during execution where the task
completed before ``TaskGraph.join()`` was called would not be raised. Now,
if a task raises an exception, its exception will always be raised when
either ``Task.join()`` and ``TaskGraph.join()`` is called.
* Fixed an issue where tasks with ``hash_algorithm='sizetimestamp'`` would,
under certain conditions, fail to re-execute when they should. This only
occurred when a graph writing the same amount of , but possibly different,
data is executed successively, with less than about 1.5 seconds between
task executions.
* After many years with the Natural Capital Project, Rich Sharp has stepped
down from the Project and as the maintainer of ``taskgraph``. James
Douglass is taking his place, and this change is now reflected in
``setup.py``.
* Fixes an issue that causes an ``EOFError`` or ``BrokenPipeError`` to occur
when the ``TaskGraph`` terminates.
* Updated the ``taskgraph`` example in the README for the latest API changes
and to clarify the need for ``if __name__ == '__main__':``
* Fixed an issue that could cause the ``TaskGraph`` object to hang if
duplicate ``Task`` objects were created.
* Fixed an issue that was causing TaskGraph to ignore a changed
``hash_algorithm`` if the TaskGraph was created on one run, was
deconstructed, then restarted. If the user chose a different hash, TaskGraph
would use the hash that the target file was originally hashed under rather
than the new algorithm.
* Removed ``copy_duplicate_artifact`` and ``hardlink_allowed`` parameters
and functionality from TaskGraph. This is to address a design error that
TaskGraph is not well suited for caching file results to avoid
recomputation. Rather than add additional complexity around the limitations
of this feature it is being removed to guide a design toward a standalone
cache library if needed.

-------------------
* Fixed issue that could cause combinatorial memory usage leading to poor
runtime or ``MemoryError`` if a dictionary were passed that had thousands
of elements.
* Fixed issue that would cause ``TaskGraph`` to not recognize a directory
that was meant to be ignored and in some cases cause ``Task`` to
unnecessarily reexecute.

-------------------
* Fixed an issue that would raise an exception when `__del__` was
deconstructing a taskgraph object and a thread ``join()`` would cause a
deadlock.

-------------------
* Fixed an issue that would ignore the state of a ``transient_run`` flag if
a previous Task run had run it with that flag set to False.
* Removed a limit on the number of times ``TaskGraph`` can attempt to update
its database up to 5 minutes of continuous failures. This is to address
expected issues when many parallel threads may compete for an update.
Relevant information about why the database update fails is logged.
* Fixed an issue where the logging queue would always report an exception
even if the logging thread shut down correctly.

-------------------
* Fixed several race conditions that could cause the ``TaskGraph`` object to
hang on an otherwise ordinary termination.
* Changed logging level to "INFO" on cases where the taskgraph was not
precalculated since it's an expected path of execution in ``TaskGraph``.
* Adding a ``hardlink_allowed`` parameter to ``add_task`` that allows the
attempt to hardlink a file in a case where a ``copy_artifact=True`` may
permit one. This will save on disk space as well as computation time
if large files are not needed to copy.
* Adding a ``store_result`` flag to ``add_task`` that conditionally stores
the ``func`` result in the database for later ``.get``. This was added to
guard against return types that were not picklable and would otherwise
cause an exception when being executed normally.
* Fixed issue that would cause the logger thread to continue reporting status
after all tasks were complete and the graph was closed.