This should be the first step towards multi-level LLM decision
making, e.g., use-cases where we have an outer loop containing e.g.
high-level decision making, and an inner loop that e.g. contains
a concrete sub-usecase execution.

For example: we have an outer loop that can iterate over multiple
attack classes, and an inner loop that executes examples for the
respective (current) attack class

This is a rough prototype that:

- inititally downloads and executes an enumeration tool (lse)
- executes and caputure its output
- use gpt-4-turbo to generate up to 3 attack classes
- use gpt-3.5-turbo with up to 20 steps per attack class to generate commands

Now we have:

- usecases/base.py for basic wiring
- usecases/common_patterns.py for reusable control loops/pattersn
- usecases/privesc/windows.py for all windows-based use-cases
- usecases/privesc/linux.py for all linux-based use-cases (migth be splitup further)
- usecases/privesc/common.py for shared code within usecases/privesc