adapt preprocessing

[AnFreTh]

This pull request includes several updates to the README.md file and new functionalities in the mambular/arch_utils directory. The most important changes include the addition of a new CNN block, enhancements to the batch ensemble layer, and the introduction of a multi-head attention module with batch ensembling.

Documentation Updates:

• README.md: Removed the "Results" section, including detailed results, average rank table, and related content.
• README.md: Corrected the description of "Periodic Linear Encoding (PLE)" to "Piecewise Linear Encodings (PLE)" in the preprocessing tools section.

New Functionalities:

• mambular/arch_utils/cnn_utils.py: Added a new CNNBlock class that allows for configurable convolutional, pooling, and dropout layers.

Enhancements to Batch Ensemble Layer:

• mambular/arch_utils/layer_utils/batch_ensemble_layer.py: Added the option for "normal" initialization in the reset_parameters method. [1] [2] [3]
• mambular/arch_utils/layer_utils/batch_ensemble_layer.py: Introduced a new MultiHeadAttentionBatchEnsemble class, which implements multi-head attention with batch ensembling.

Minor Configurations:

• mambular/arch_utils/layer_utils/embedding_layer.py: Changed the default embedding_type to "linear" and added an embedding_bias attribute.