Include new models and major restructuring in util layers

[AnFreTh]

This pull request includes several significant updates to the Mambular package, including additions to the README, new custom training examples, enhancements to the normalization layer, and new batch ensemble layers. The most important changes are summarized below:

New Models included.

• Added new models TabM, NODE, and BatchTabRNN.

Documentation Enhancements:

• Added descriptions for new models TabM, NODE, and BatchTabRNN in the README.md file.
• Introduced a section on custom training in the README.md, providing an example of how to implement a custom training loop using mambular.base_models.

Code Enhancements:

• Added support for None normalization layer in the get_norm_fn function.

New Features:

• Implemented LinearBatchEnsembleLayer and RNNBatchEnsembleLayer classes in mambular/arch_utils/layer_utils/batch_ensemble_layer.py to support batch ensembling for linear and RNN layers.

Refactoring:

• Refactored the EmbeddingLayer class in mambular/arch_utils/layer_utils/embedding_layer.py to use a configuration object for initialization, simplifying the constructor and making it more flexible. [1] [2]
• Improved the forward method of the EmbeddingLayer class by adding comments for clarity and restructuring the code for better readability. [1] [2] [3] [4] [5]