Add more debug info so fixing YAML files and wrong paths is easier
Add a send_as and ai_name setting so when conversation strings are built, some models can follow the AI conversation better
If models are behaving, they send the plan attribute of the response as a list, which is correct given the YAML, but Auto-GPT want's a YAML string, not an array. Auto-GPT will now always get a YAML bulleted list back.
Added strip_messages_from_end because when an RP prompt is sent, the last message (or messages) can confuse the AI into taking the same course of action. Changing it from zero to an integer allows people to strip the "Determine your command..." message, and possibly the time message which takes up pointless space since the AI has no concept of time anyway.