Switch to using the more powerful GPT-3.5 LLM via OpenAI

[hardbyte]

As mentioned in #5, the recommendations suggested by this project were just not very good. Also the codex model we are currently using is apparently being deprecated tomorrow - not that I believe it will actually stop working straight away. https://news.ycombinator.com/item?id=35242069

This PR replaces codex with gpt-3.5-turbo and uses their official Python wrapper instead of directly calling the rest API.

[giorgiop]

GPT-4:

This pull request addresses the issue mentioned in #5, where the recommendations provided by the project were not very good. The proposed changes switch from the Codex model to GPT-3.5-turbo and use OpenAI's official Python wrapper instead of directly calling the REST API.

The changes made are reasonable, and it's likely that GPT-3.5-turbo will provide better recommendations. I've reviewed the code changes and they look good. I suggest approving this pull request.

Please make sure to test the updated code to ensure it works as expected before merging the changes.