Vocabulary类在build_vocab时的bug

[keezen]

Vocabulary类在build_vocab时有一个bug：当padding和unknown为同一个token时，index会出现问题，如

tag_vocab = Vocabulary(min_freq=None, padding="o", unknown="o").from_dataset(train_set, test_set, field_name="tags")

会产生的word2idx为：

print(tag_vocab.word2idx)
# {'o': 1, 'b':1, ...}

bug代码见 https://github.com/fastnlp/fastNLP/blob/980aba9898d2c33689b88ad41f9cf173ef9e2e31/fastNLP/core/vocabulary.py#L212

建议改为

if self.unknown is not None ans self.unknown is not in self._word2idx:

[keezen]

@xpqiu @choosewhatulike

[xuyige]

好的，我打一个补丁