Refactored to use typed dicts. Removed stale code

[C-Loftus]

Kept same behavior the same as far as I can tell.

• Removed all dict with any as a key or value. Try to use GPTMessage or GPTMessageItem as the type as much as possible. Let me know if there is a name you prefer instead of these. I wasn't sure what to name them.
• Got rid of old tts and image description code that I don't think is relevant anymore. These don't follow the grammar and aren't being used since they were broken but no one reported an issue
• No pylance errors are being thrown under standard type checking
• Got rid of the a11y editor context logic in the system prompt. I will probably add something to make that more explicit. I don't like that it would be always passed in if the user has screen reader support enabled for axkit
• Use the word {user.model} **start** thread: since it is one syllable instead of enable which is 3 syllables.
• Changed contextual_user_context to be gpt_additional_user_context since it is a globally exposed talon python function, we want to prefix it with gpt. I personally find additional_user_context to be more clear than contextual_user_context which strikes me as a bit confusing.

I am using TypeDict so we keep all the methods within the standard dict class. Was considering creating a super class for the message and inheriting from it but I think this abstraction does not actually benefit and adds boilerplate.

[C-Loftus]

Have not tested this yet. Feel free to test if you want. Going to take a break for a bit.

[C-Loftus]

Believe this should be good now. Tested with threading / context / images etc, and doesn't seem to have any regressions. The threading is very cool thank you for your work on this. I find it works well. I will add funcitonality to have the imgui window auto-update with a new thread response once this merged.

[jaresty]

I'll use this branch for a while. Looks good though!