* Add tutorial about nlp dataset customization

* Update doc and tokenizer usage in seq-cls demo

* Updata tutorial