It seems that you are getting two problems with your pipeline:
The item_list of the second extractor should be the state[‘texts’] (that is, the email you are classifying instead of “item”
The HTML to text extractor returns an array of paragraphs, so the keyword extractor is getting an array as the input and that is why is returning an error.
In order to solve this problem, you should join all of the paragraphs returned into one single emailtext as shown in the “transform” part of the Pipelines documentation. Then, that unique text will be the input of the keyword extractor.