Extract dataset functionality for easy extensibility

[alxkolm]

Extract dataset functionality for easy extensibility

Summary of changes:

1. Added dataset.py with Dataset base class. It encapsulates downloading and iterating over examples in files. There are 3 methods download(), list_files(), examples_of() in the class
2. Download functionality moved from tinystories.py to Dataset
3. These functions now receive Dataset as argument: train_vocab(), pretokenize(), process_shard()
4. Pre-tokenized files now write to tokenized_{vocab_size} directories. Files tokenized by Llama2 tokenizer write to tokenized_llama2 directory.
5. Wrapped Tinystories dataset to TinyStories class in dataset.py
6. Added new SQLCreateContext from sql-create-context as example of extensibility.

To hold git diff as simple as possible, I have not renamed the main entry point file tinystories.py. But in future, it should be renamed to something like prepare.py.

train.py is not affected by this PR.