Cache get_event_loop()'s per-thread loop instead of re-resolving every call

[wbarnha]

Description

get_event_loop() fell all the way through to asyncio.get_event_loop() (raise-then-catch RuntimeError, then new_event_loop() + set_event_loop()) on every single call once there was no running loop — the common case for calls made outside of a running loop, e.g. at import time when services/agents are declared at module level.

Cache the resolved loop in thread-local storage and reuse it on later calls from the same thread, skipping the asyncio.get_event_loop() dance entirely once a loop has been resolved. asyncio.get_running_loop() is still checked unconditionally on every call and can't be cached — whether a loop is currently running changes on every call by definition — but that's the cheap path (no exception when a loop is actually running); the expensive not-running path is what benefits from caching.

The cache is thread-local (threading.local()), not a single global/module-level singleton: ServiceThread runs a dedicated event loop per worker thread (mode/threads.py), including code that compares get_event_loop()'s result by identity against a specific thread's loop (if get_event_loop() is self.parent_loop). A plain global would return the wrong thread's loop across worker threads.

Verification

• Added tests: caching skips a repeated asyncio.get_event_loop() call, and a second thread's cached loop is independent of (and doesn't leak into) the main thread's.
• Full suite: 740 passed on Python 3.11 and 3.13, ruff clean. Loop and thread-related tests re-run repeatedly with --random-order to rule out cache-staleness order-dependence.
• Cross-checked against a real faust-streaming/faust checkout (same setup used to verify Fix two annotation-resolution regressions from the 3.13 rewrite #75): full tests/unit + tests/functional — 2128 passed, 32 skipped, 0 failed, unchanged from before this change.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HgnKFtXZbCjXNoVNWa5JLd

---

Generated by Claude Code