Solved another spec race condition with that effects JRuby

[patvice]

Problem
JRuby CI was intermittently failing in Cancellation Integration (stdio-native) due to concurrency races in cancellation paths. The failing spec relied on unsynchronized shared state between threads, which is more likely to flake on JRuby / cause issues in production settings though MRI should be safe to do it's thread model.

I also found a related runtime race where a cancellable operation could be marked running before its worker thread was safely published, causing cancellation to miss the thread in rare timing windows.

Solution

• Made cancellation integration specs thread-safe by replacing shared mutable variables/arrays with Queue-based signaling.
• Updated cancellation worker thread handling to tolerate TransportError during teardown races.
• Fixed CancellableOperation to publish state and thread atomically under mutex, so cancellation always sees a consistent state.
• Hardened concurrent state/value reads in Promise and AsyncResponse with synchronized accessors.
• Added a small timeout helper cleanup improvement (kill + short join) to reduce orphaned worker threads after timeout.