* Also fix some other bugs.
* Fix #1495

* Test gzip

example usage:

import paddle.v2 as paddle

if __name__ == '__main__':
    dict_en, dict_fr =	paddle.dataset.wmt14.build_dict()
    train = paddle.dataset.wmt14.train(dict_en, dict_fr)
    test = paddle.dataset.wmt14.test(dict_en, dict_fr)

    total_train = 0
    for i in train():
        total_train += 1

    total_test = 0
    for i in test():
        total_test += 1

    print total_train, total_test

remove inferencer and Inference from __All__, since infer is sufficent, they are implementation detail.

rename inferencer.py to inference.py

* Change build process to generate v2 documentation
* Add directory for v2 api
* Correct sphinx index soft-link

I use pre-commit check all file and report is the passed.
But it is not passed in Travis, so I update __init__.py

remove ITrainer interface since parameters for train is already getting out of sync. Currently there is no benefit of having a interface.

expose build_dict in imikolov dataset, fix bug that len(word_dict) is not bigger than all index in word_dict.