test(timed_cache): de-flake test_expiry_removal (#1408)

[yashksaini-coder]

Summary

Closes #1408.

test_expiry_removal fails intermittently on CI (seen on windows (3.11, core); green on all Linux runners and locally):

assert MSG_1 not in cache.cache
AssertionError: assert b'msg1' not in {b'msg1': <expiry-ts>}

Root cause

cache = LastSeenCache(ttl=2, sweep_interval=1)
cache.add(MSG_1)
await trio.sleep(2.1)  # Wait for sweeper to remove expired items
assert MSG_1 not in cache.cache

BaseTimedCache sweeps in a background thread (while not self._stop_event.wait(self.sweep_interval)) with an integer-second TTL. Removal therefore happens on a coarse ~1s cadence in a separate thread. The fixed trio.sleep(2.1) assumes the sweep tick lands within 2.1s; on a slow Windows runner the thread wakeup is delayed and the entry is still present when the assertion runs. A timing race, not a logic bug.

Fix

Poll until the sweeper removes the entry — the idiom test_simple_last_seen_cache already uses in this same file:

-await trio.sleep(2.1)  # Wait for sweeper to remove expired items
-assert MSG_1 not in cache.cache  # Should be removed
+with trio.fail_after(15):
+    while MSG_1 in cache.cache:
+        await trio.sleep(0.1)

Deterministic (exits as soon as the entry is swept) and non-hanging (fails after 15s if the sweeper is genuinely broken).

Scope

Limited to the test that actually failed in CI. Sibling tests in the file (test_simple_first_seen_cache, test_timed_cache_expiry, test_timed_cache_stress_test, test_readding_after_expiry) share the same fixed-sleep pattern and are latent flakes on the same root cause — notable as a follow-up, out of scope here.

Test plan

• test_expiry_removal run 5× locally → stable pass
• Full tests/core/tools/timed_cache/ → 8 passed
• ruff check / ruff format --check clean
• newsfragments/1408.internal.rst added