fix: Fixed the issue where the mii.pipeline.pipe(stop) was ineffective

[kitstar]

1. Replace Tokenizer.encode with Tokenizer.convert_tokens_to_ids
2. Support multi stop tokens and batch inference

Easy reproduce like
responses = pipe(args.prompt, do_sample=False, max_new_tokens=args.max_new_tokens, return_full_text=True, stop=["<|im_end|>", "<filename>"])

[kitstar]

@microsoft-github-policy-service agree

[mrwyattii]

Thank you @kitstar - this has been something on my TODO list that I was not able to carve out time for. I'll get this reviewed and merged ASAP!