Refactorization

[AnFreTh]

This pull request includes several changes to the mambular package, focusing on refactoring imports and enhancing the functionality of the BaseModel and TaskModel classes. The most important changes include moving BaseModel and TaskModel to a new utils directory, updating import paths accordingly, and adding new methods for pretraining embeddings.

Refactoring imports:

• Moved BaseModel and TaskModel to mambular/base_models/utils/ and updated import paths in all relevant files. [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13]

Include new models

• New model architectures TromptandAutoIntare now included into mambular.

Introduce pretrainingoption:

• Added pertaining functionality for all models that use encoders.

Include dilation to 1DConv layers in Mambular/TabulaRNN

• See Issue Add dilation in convolution #182

Enhancements to BaseModel:

• Added new methods for embedding management, including embedding_parameters, encode_features, get_embedding_state_dict, and load_embedding_state_dict.
• Modified the encode method to support gradient computation and shuffling embeddings.

Enhancements to TaskModel:

• Added a new method pretrain_embeddings for pretraining embeddings using contrastive learning.
• Introduced helper methods get_knn and contrastive_loss to support the new pretraining functionality.

Miscellaneous changes:

• Added print statements for debugging and logging purposes. [1] [2]
• Minor formatting changes for code readability.