update JanomeSubwordsTokenizer for transformers>=4.34

[mh-northlander]

fix #335.

From transformers>=4.34, PreTrainedTokenizer.__init__ requires self.vocab to be set.
Move the super(...).__init__ call to the end of JanomeSubwordsTokenizer.__init__ (and use unk_token instead of self.unk_token before init), following changes of BertTokenizer at that time.

Also move the call of self.add_tokens since it requires super(...).__init__ is done.