fix: tps not initialized with 0, skip_special_tokens not exists in TextStreamer options

[nekomeowww]

Summary

I realized several issues when trying to replicate the Whisper WebGPU example in my development environment:

• The event of update may contain undefined value for tps, and with the semantic of the code, tps is used to calculate the speed of outputting tokens, where it should be consistant as number since it works like metrics.
• According to Transformers.js documentation, instead, skip_special_tokens should be PreTrainedTokenizer.decode(...)'s option.

Changes

• Initialize tps with 0.
• Move skip_special_tokens into decode_kwargs as one of the property.
• Added // eslint-disable-next-line no-unused-vars comment to disable lint check for line const [tokenizer, processor, model] = await AutomaticSpeechRecognitionPipeline.getInstance(x => { or lint will fail.