Implement encoder-only T5 sequence classification

[zeynepyirmibes]

Normally T5forSequenceClassification loaded the encoder and the decoder, but this is unnecessary for text classification tasks. To functionally use the "classification" task format, I implemented a T5 classifier that only loads the encoder (taking help from @gokceuludogan's wrapper).

New:

• T5forSequenceClassification classifier that only loads the encoder
• Specific tokenization for "classification" task format
• Ability to train with other sequence classifiers, such as BERT models
• Regression mode activated for semantic similarity, so that STS can be performed with only the encoder

[gokceuludogan]

• Implemented NER preprocessing and postprocessing for classification mode.

• Adjustments were made to the dataset_preprocessor, trainer, and evaluator.