Correct and clean up JSON handling (Significant-Gravitas#4655)

* Correct and clean up JSON handling

* Use ast for message history too

* Lint

* Add comments explaining why we use literal_eval

* Add descriptions to llm_response_format schema

* Parse responses in code blocks

* Be more careful when parsing in code blocks

* Lint

autogpts/autogpt/tests/challenges/current_score.json

@@ -53,4 +53,4 @@
             "max_level_beaten": null
         }
     }
-}
+}