Fix inference for Vicuna v0 weights

[laidybug]

I modified the conversation.py file to fix the accidental removal of the logic for the v0 weights. Using v0 stopped functioning correctly after the introduction of v1.1.

Now both versions work properly, with the expected model paths being similar to Vicuna-7B-v0 and Vicuna-13B-v1.1 so v0 can be detected. Issues https://github.com/lm-sys/FastChat/issues/409 and https://github.com/lm-sys/FastChat/issues/422 are likely related to the omission this contribution fixes.