v0.2 with new v2 model support

this version updated the code to suit index-tts v2 and hopefully fixed bugs.
v2/generate endpoint is tested, anything else isn't.
known issue: each task batch will take about 3 minutes when used with a gpu on preparation, that's normal and is something they do inside index-tts, there's nothing much i can do about it.