Elephas

- Adding support for training, inference, and generation Hugging Face Causal LM models (https://github.com/danielenricocahall/elephas/pull/39).

- Support Hugging Face models (Tensorflow backend) for classification models and synchronous training onnly (use `pip install elephas[huggingface]`).
- Updated various dependencies
- Removed Support for Python 3.8 due to EOL in October 2024

- https://github.com/danielenricocahall/elephas/pull/32: Raise an exception with a clear error message for developers using Tensorflow 2.13+ and an optimizer which is not a subclass of the `Optimizer` class from `tensorflow.keras.optimizers.legacy` (e.g; subclasses of `tensorflow.keras.optimizers`), as non-legacy optimizers cannot currently be serialized

- Add support for distributed inference and evaluation of multiple input/output models: https://github.com/danielenricocahall/elephas/pull/34

- Added multiple input/multiple output training capabilities for `SparkModel`: https://github.com/danielenricocahall/elephas/pull/31

- Update to support Python 3.11 and consequently later Tensorflow versions (>2.10). This required changing the optimizer imports in the unit tests to use the legacy optimizer (`tf.keras.optimizers.legacy`), as detailed in [the corresponding PR](https://github.com/danielenricocahall/elephas/pull/25). For developers using a version of Tensorflow <2.11, importing from the optimizer package (`tf.keras.optimizers`) will still function as expected. Otherwise, the imports will need to be changed to the legacy optimizer to ensure your application doesn't break.