Make retry queue durable across transient delivery failures

[marandaneto]

Summary

Make the Ruby SDK retry queue durable across transient delivery failures.

Spec

• sdk-specs/openspec/specs/retry-queue/spec.md

Current evidence

The SDK audit found that SendWorker removes events from the queue before send and does not requeue on non-200/transient failure.

Expected behavior

• Keep or requeue events after retryable/transient delivery failures.
• Deliver retained events after a later success.
• Bound retained events by queue capacity.
• Drop non-retryable failures appropriately.
• Add tests for 503/network failure retention and later delivery.