[Feature] Add SentencePieceTokenizer for BERT (#669)

* add BERTSPTokenizer

* update doc

* fix doc

* add BERTVocab.from_sentencepiece()
add test case

* fix doctest

* fix lint error

* fix doctest and lint error

* update for unicode processing(py2)

* remove comments

* disable lint error

* fix flake8 error

* make vocab consistent

* remove six. make convert2unicode private

Author: haven-jeon