Bugfix: Fix several issues relating to GPT4

[Hialus]

Pyris was having severe issues in production environments when trying to use GPT4. I've been debugging this over several weeks and found 3 separate issues at play:

• A while back we added some caching for guidance objects, which guidance did not properly support in concurrent contexts -> Removed
• Guidance also does not like it if there are concurrent requests en masse, like the health check produces them -> Added a lock for the health check
• The OpenAI API returns None as the message for empty messages. Guidance did not support this -> Manually overrode guidance function

[MichaelOwenDyer]

Thank you for fixing this!!