Retrying

We introduce a retry argument on the @ai decorator to make LLM execution a little more robust. We also fix a few type checking errors.