This fixes two small bugs that were unlikely to be encountered but present:
if the code interpreter was passed as a tool to a run (rather than an assistant), the openai API fails to recognize and cast it, resulting in an API error
assistant instructions were not provided with every message. This matters for Applications, where the instructions contain the latest state object (so they need to be constantly refreshed). In practice, this would only be experienced by users that were using manual lifecycle management; with lazy lifecycles, the assistant (and instructions) are created on every call, meaning the bug couldn't impact.
This fixes two small bugs that were unlikely to be encountered but present:
Applications, where the instructions contain the latest state object (so they need to be constantly refreshed). In practice, this would only be experienced by users that were using manual lifecycle management; with lazy lifecycles, the assistant (and instructions) are created on every call, meaning the bug couldn't impact.