Fix `UnboundLocalError` and make model configurable

[jcp]

This commit addresses two issues in software/source/server/i.py:

1. Fix the UnboundLocalError caused by importing the os module at both the top level and within the configure_interpreter function.
2. Make the interpreter.llm.model configurable via the MODEL environment variable or --model command line argument. Defaults to gpt-4. This allows users to run O1 locally.

Fixes #116

[HanClinto]

Looks great to me!