Test suite fails against guard-core 3.15.0 (redis_fail_open semantics) #126

Value this delivers

CI is green on master again with guard-core 3.15.0, and the suite asserts the 3.15.0 Redis-failure semantics explicitly (fail-secure by default, in-memory fallback only with redis_fail_open=True), so the next guard-core change in this area is caught instead of masked.

What happens

Since guard-core 3.15.0 (PyPI 2026-08-27 15:52Z) test_rate_limiter_redis_errors raises GuardRedisError and the two Redis websocket tests close with 1013, on every CI run (3 failed, 381 passed). The last green master run installed 3.14.0. The tests encode the 3.14.0 silent fallback; in the websocket tests a Redis-handler event-loop mismatch that 3.14.0 swallowed now surfaces.

Acceptance

Full suite green locally and in CI against guard-core 3.15.0; the three tests assert the intended behaviour under both redis_fail_open values where applicable; no assertion weakened, no skips.