Wolframclient

- WXF deserialization maps `List` to `tuple`, was `list` before for two reasons. `List` and tuple are immutable. `tuple` contrary to `list` are hashable object and as such can be keys in `dict`.
- WXF deserialization maps numeric arrays to `NumPy` arrays. Numpy being a pre-requisite of the library, and lazily loaded, this change will boost performances while causing very little trouble, if any.
- Fix bug in `WolframLanguageSession` that was preventing Python interpreter from exiting.
- Add new class `PackedArray`, a wrapper on top of NumPy `ndarray`. Instances of `PackedArray` are serialized as... WL packed array (!). This fixes the issue of having a WL packed array (say the output of `Range[3]`), being serialized to a `ndarray` (`arange(3)`), and send back to the kernel as a `NumericArray`.
- Use `black` code formatter.

- Add two new optional flags to `test` command called `-x` or `--xml` and `-d` or `--xml-dir`. It produces an xml output of the test results using the `unittest-xml-reporting` library. `-d` expects the filepath of the output directory, default is `test-reports`. Also add `-v` or `--verbosity` to control the `verbosity` option of the test suite. Check `python run.py test -h` for more info.

- Bug fix in `WXFConsumer`.
- Update some URLs.

- Introducing a new class `WolframKernelController`. Major code change to ensure ZMQ sockets are isolated in a given thread. Add a new controller class, a `Thread`, that own the sockets connected to the kernel. Interaction with this thread is done through a Queue.
- Merge `WolframLanguageFutureSession` into `WolframLanguageSession`. All evaluate methods have a future counterpart.
- Merge `WolframCloudFutureSession` into `WolframCloudSession`. Add relevant future methods.
- Rename `WolframAPICall` method `add_parameter` to `set_parameter` since setting is exactly what it does.
- Result of API call, the `WolframAPIResponse` family of classes, now automatically deserialize the response content using the content-type, if it's `WXF` or `JSON`. The content type remains accessible in the `response` member of those classes.
- rename evaluator parameters `STARTUP_READ_TIMEOUT` to `STARTUP_TIMEOUT`, and `TERMINATE_READ_TIMEOUT` to `TERMINATE_TIMEOUT`.
- CPU consumption reduced to ~1% with the latest ZMQLink version.
- Local session checks for the Wolfram Kernel version at startup and fails with a specific message.
- Local session supports instant logging for messages, with severity `warn`, and prints, with severity `info`. This only works for kernel with a `kernel_loglevel` set on initialization, and allows real time feedbacks, even if the kernel evaluation is not yet completed.
- Change mapping of WXF symbols to Python objects. The only symbols that are deserialized to Python objects are the one that round trips: `Null`, `True`, `False`. `Pi` and `None` are no more deserialized to, respectively, `math.pi` and `None`.

- Inconsistent class name `WolframCloudSessionFuture` replaced with `WolframCloudFutureSession`.
- Rework completely `Dispatch` class. Add a `Dispatch` instance called `wolfram_encoder` representing the multi dispatch function that encodes Python object.
- Add an entry point allowing new encoders to be registered as separated libraries (plugins).
- Add support for pandas classes: `Series` and `DataFrame`.
- Extend support for NumPy numbers.
- Local evaluators can be initialized without a Wolfram Kernel path, in which case default paths are scanned to find the most recent product.

- Add `CHANGELOG.md` to the repository to help users track the changes.
- Major refactoring of the evaluation module.
- add base classes for synchronous and asynchronous evaluation, specifying the evaluation scheme.
- add cloud session implementation based on event loop using the aiohttp library.
- rename asynchronous session based on future module from `WolframCloudAsyncSession` to `WolframCloudSessionFuture`.
- add new implementation of `WolframCloudAsyncSession` based on asyncio and coroutines.
- Cloud session named option `authentication` renamed `credentials`.
- Significant improvement of `PIL.Image` support. Speed up serialization and deserialization of images, and add support for more input formats.