Data generation / retrieval for Training

[JakubSchwenkbeck]

Motivation

Implemted a fully scalable and reliable mechanism to retrieve datasets from a raw text and tokenize it. Also added new structs which hold all values (matrices) concerned by the training/loss optimization
Resolves #13

Changes

• Added a new generation.rs file which can read a raw text file and process the data, then generate a input and target pair for training
• tokenizer is reworked to have some more public functions and only need to take the raw input texts as sentences (vec<&str>)

Test

Run with an example of choice and observe outputs