test: fill signal, sandbox, and TaskWrapper edge-case stubs

[skundu42]

Summary

• Implemented edge-case coverage for concurrent signals during generation (test_signal_edge_cases.py)
• Implemented sandboxed CodeAct REPL edge cases: state mutation, tool brokering, and execute_python NameError recovery (test_sandbox_edge_cases.py)
• Added TaskWrapper contract tests for task.abort / task.replan / task.current, skipped until that builtin is implemented (test_task_wrapper_edge_cases.py)

Test plan

• uv run pytest tests/edge_cases/test_signal_edge_cases.py tests/edge_cases/test_sandbox_edge_cases.py tests/edge_cases/test_task_wrapper_edge_cases.py -v

[alessiodevoto]

Hi @skundu42! Thanks for opening this PR 🙂

I ran locally and got 7 passed, 3 skipped, I noticed the TaskWrapper tests but we actually don't have that implemented yet, so I’m not comfortable merging the TaskWrapper part as task.abort, task.replan, and task.current are not implemented today. Please remove tests/edge_cases/test_task_wrapper_edge_cases.py.

For the signal tests, I think those make sense but could you please make the lock assertion deterministic, for example replace the asyncio.sleep(0.05) checks with an explicit event/counter in the new GatedFakeLLM, and assert that the nested generation has not entered the LLM call before the outer generation releases the lock ? I left a comment in the code too!

[skundu42]

@alessiodevoto Thanks for the review. I have made the requested changes. Let me know if this works.

Also, please feel free to assign me other issues directly. I am planning to be an active contributor in this.

[alessiodevoto]

Hi @skundu42 thanks for updating ! LGTM, merging 🙂 !