Fix parse_input

[gAbelli]

Fixes #11 (I believe)

Correct me if I'm wrong, I don't fully understand why the function is parsing and returning args in the first place.
With this change there are no problems in interactive mode, but I don't know if this causes problems elsewhere.

[kharvd]

The reason why we parse args from the prompt is to support overrides like

> This prompt will be run with a different model --model gpt-4

(inspired by Midjourney)

I use this a lot, but can see how it might be annoying if you're pasting some command-line args to the prompt. Not sure what the right solution is, maybe some kind of escaping?

[sghael]

I hit this bug quite often and I'm assuming many others do. It's not uncommon to cut and paste terminal output into the prompt, to inquire the LLM for help in debugging. Those debugging traces will often have --.

I believe the correct solution is to allow users to triple quote or triple backtick large blocks of input. Triple quoted text should be taken literally into the context, and not parsed.

> This prompt will be run with a different model --model gpt-4
> ```This prompt will be run with a different model``` --model gpt-4

> ```This prompt will NOT be run with a different model --model gpt-4```
> """This prompt will NOT be run with a different model --model gpt-4"""