Add a new option to allow manually setting encoding

[imgaojun]

resolve #270
I can not load utf-8 file while building my vocabulary or loading my dataset because gbk is used by default on windows. I added a new option to allow manually setting encoding PairedTextData

$ python main.py 
Traceback (most recent call last):
  File "main.py", line 62, in <module>
    main()
  File "main.py", line 28, in main
    hparams=config_data.train, device=device)
  File "C:\Users\gaojun4ever\Miniconda3\lib\site-packages\texar\torch\data\data\paired_text_data.py", line 140, in __init__
    eos_token=src_hparams.eos_token)
  File "C:\Users\gaojun4ever\Miniconda3\lib\site-packages\texar\torch\data\vocabulary.py", line 103, in __init__
    = self.load(self._filename)
  File "C:\Users\gaojun4ever\Miniconda3\lib\site-packages\texar\torch\data\vocabulary.py", line 119, in load
    vocab = list(line.strip() for line in vocab_file)
  File "C:\Users\gaojun4ever\Miniconda3\lib\site-packages\texar\torch\data\vocabulary.py", line 119, in <genexpr>
    vocab = list(line.strip() for line in vocab_file)
UnicodeDecodeError: 'gbk' codec can't decode byte 0x8c in position 2: illegal multibyte sequence

[gpengzhi]

Could you fix the CI error? CI is currently failing because the modified lines of code are too long (> 80). And if we want to change the code, we also need to update the corresponding documentation accordingly.

[huzecong]

I guess we could add encoding arguments for functions where we perform file I/O.

Another question is, would it be helpful to set the default to UTF-8, regardless of platform and locale? IIRC the default encoding on Windows is locale dependent (e.g. GBK for Chinese systems).