Enable split LLM usage between Intro post and summary replies

[remlaps]

For example:

• It would probably be useful to use a free model for the summary replies and a paid model for the intro post.
• It would probably be useful to do multiple screenings with free models before a final pass with a paid model.

This would the ability to pick from a list for the following variables, "on the fly"

• ARLIAI_MODEL
• ARLIAI_URL
• LLMAPIKEY

Also, that "ARLIAI" naming is outdated. It should be changed (for all relevant variables) to something that's not specific to any LLM provider.