[InputData] Support texts as InputData

Impement experimental pipeline with simple NLP-model (or even NaiveBayesClassifier) for text classification/sentiment analysis:

New InputData format with generator-based logic (for texts batch-loading)
New simple model (for instance, CNN) or adoptation the existing models that are suitable for text classification
Maybe bag-of-words, word-2-vec model, etc. for text embeddings
Several text preprocessing methods (stemming, lemmatization, filtering, etc.)
Example of dataset where pipeline can be applied (spam classification for instance as the simpliest case)
Example of FEDOT pipeline that can solve the given task

aimclub / FEDOT