Improved text generation & streamed responses

I added a new system prompt for your cool text generation (when no text is selected) feature, as it previously used the Describe your change prompt and could confuse the LLM.

I optimised time.sleep() make the streamed responses work faster (to the extent that it won't break!)