fix(open-interpreter): terminate disconnected stream workers

[tang-vu]

Summary

• terminate the Open Interpreter streaming subprocess when a client disconnects
• escalate to kill() if the child ignores the bounded termination wait
• close subprocess pipes and remove the temporary runner on every exit path
• keep SSE payload framing stable while removing subprocess newlines

Why this matters

POST /chat/stream starts a Python worker for every stream. When a browser, proxy, or client disconnects after receiving a chunk, Starlette closes the generator, but the old implementation only removed the runner file and left the worker alive. Repeated disconnects could accumulate interpreter processes, CPU/memory use, and in-flight model work until the optional service became unhealthy.

Overlap check

Searched open and closed upstream PRs for open interpreter stream subprocess cleanup, open-interpreter, disconnect, and the changed production file extensions/library/services/open-interpreter/server.py. PR #403 hardened authentication and container execution but does not manage streaming child lifecycles. No open or closed PR covers disconnect cleanup.

Test plan

• python -m pytest -q ods/extensions/library/services/open-interpreter/tests/test_server.py (3 passed)
• python -m py_compile ods/extensions/library/services/open-interpreter/server.py

The regression tests exercise the stream generator boundary: closing after the first SSE event terminates the live worker and removes the runner; normal completion does not terminate an exited worker; an unresponsive worker is killed.

Generated with Codex.