Fault-tolerance primitives — retries, circuit breakers, timeouts, fallbacks, and health checks

Failure modes

Symptom	Cause	Fix	Severity
`with_fallback(...)()` returns a coroutine / `RuntimeWarning: never awaited`	`with_fallback` returns an async wrapper	`await` it (or `asyncio.run`); or use the sync `@fallback` decorator	high
Calls keep raising `CircuitOpenError`	breaker is OPEN after `failure_threshold` failures	wait `reset_timeout`, or `get_circuit_breaker(name).reset()`	medium
Retries never stop / take too long	`max_attempts` / `max_delay` too high	tune the knobs; set `retryable_exceptions` to narrow what retries	medium
Every exception retries, even bugs	`retryable_exceptions` left as default (all)	pass the specific exception tuple	medium

with_fallback / with_timeout are async. The @fallback decorator is synchronous on a sync target; with_fallback returns an async wrapper regardless.
Breaker state is shared by name. Two decorators with the same name share one breaker via the registry.
Unbounded retryable_exceptions. Defaulting to all exceptions retries programming errors too.

Coroutine-not-awaited? You used an async primitive (with_fallback/ with_timeout) synchronously.
CircuitOpenError? Inspect get_circuit_breaker(name).get_stats().
Slow retries? Check max_attempts / max_delay / backoff_factor.