Implement a transform method in CrossValCurate

[sumanthprabhu]

Is your feature request related to a problem? Please describe.
Currently, we support a limited set of options for curate_feature_extractor (TfidfVectorizer, CountVectorizer, SentenceTransformer) and curate_model (Sklearn models which implement predict_proba). Domain specific feature extraction methods / SOTA models are not included.

Describe the solution you'd like
If we would like to experiment with a wider set of feature extraction methods / classification models, then decoupling the cross validation based training from the label quality assessment would be helpful. Essentially, in addition to the fit_transform method, we implement a transform method that accepts the results from cross validation based training as input and performs the label quality checks. Following is an example snippet of how transform would potentially work -

crossval_pred_probability_matrix = CustomCrossValitidationTraining(CustomModel, data_with_noisy_labels)
cvc = CrossValCurate(random_state=seed, correctness_threshold=0.0)
train_data_modified = cvc.transform(crossval_pred_probability_matrix, train_data, y_col_name="label") 

• CustomCrossValitidationTraining is a custom trainer and CustomModel is a custom model both defined by the user outside of DQC Toolkit's scope.
• crossval_pred_probability_matrix contains the cross validation based prediction probabilities for each label for each sample.
• train_data_modified is the result similar to what is observed for CrossValCurate.fit_transform