chore: require circuit-breaker-box>=0.3.9 for non-blocking retries

[lesnik512]

What

Require circuit-breaker-box>=0.3.9,<0.4.

-    "circuit-breaker-box",
+    "circuit-breaker-box>=0.3.9,<0.4",

Why

The dependency was previously unconstrained, so a resolver was free to pick any version — including the releases that stall the event loop.

Every circuit-breaker-box before 0.3.9 drove its retry loops with the synchronous tenacity.Retrying, which sleeps via time.sleep(). That blocks the whole event loop for the length of each retry wait, not just the calling task. base-client exposes the affected Retrier directly through BaseClient.retrier, so any client configured with a positive wait strategy inherited the stall, as did anyone using CircuitBreakerRedis.

0.3.9 fixes all three call sites — community-of-python/circuit-breaker-box#10 for Retrier.retry() and community-of-python/circuit-breaker-box#11 for CircuitBreakerRedis.increment_failures_count / is_host_available.

The upper bound keeps the pin inside the current minor, since Retrier is re-exported as part of this package's surface.

No code change needed

base_client only touches the public Retrier API, which 0.3.9 left untouched — same name, same signature, no AsyncRetrier introduced, and attempt counts, retry predicates, reraise, callbacks and circuit breaker interaction order all preserved. The only difference is that retry() now yields during its waits instead of blocking.

Verification

Resolved and tested against the real 0.3.9 from PyPI, confirming the installed artifact carries all three fixes rather than a locally built stand-in.

pytest (62 passed, 100% coverage), mypy --strict, ruff format --check and ruff check all clean.

Second commit

chore: ignore CPY001 is separable from the bump and touches only pyproject.toml. ruff is unpinned and just install runs uv lock --upgrade, so CI resolves the newest ruff on every run; a release since the last green build on main (August 2025) enabled CPY001 under select = ["ALL"], and it reports all 13 files in the project.

This is unrelated to the bump — checking out an untouched main and running ruff check reproduces the same 13 errors, so main fails lint today as well. Same fix as circuit-breaker-box#10. Drop that commit if you would rather handle the lint drift separately.