Update tokenizer.py to fix erroneous Javanese language code

[KathyReid]

The ISO-639-1 code for Javanese is jv NOT jw as given here. It should be listed as jv.

This is a breaking change - anyone who calls transcribe() with language='jw' will get an error.

[ryanheise]

Unfortunately you can't simply fix it by renaming it in the dictionary, since the model was trained to associate jw with Javanese speech, and will only recognise Javanese speech with the jw code. So you need to pass in jw or Javanese. I suppose if you really wanted to correct for this, you could make it work by adding a mapping from jv to jw in TO_LANGUAGE_CODE although that's a hack since that dictionary is meant to map language NAMES to code.