[#695] Fix race in TraditionalWorkQueue.isIdle()

[vharseko]

Fixes #695

Problem

TraditionalWorkQueue.isIdle() checked that the operation queue is empty and that no worker thread is active (operation != null). A worker thread is invisible to both checks in the window between taking an operation from the queue (opQueue.poll() inside retryNextOperation) and assigning it to its operation field in TraditionalWorkerThread.run(). waitUntilIdle() polls isIdle() every millisecond, so it could return immediately while a just-submitted operation was still being handed off to a worker.

This made TraditionalWorkQueueTestCase.testWaitUntilIdleSlowOpInProgress (from the slow group, excluded from CI) flaky: the test submits an operation delayed by 5 seconds and expects waitUntilIdle(10000) to block for at least 4 seconds, but it occasionally returned right away.

Fix

Track the number of accepted-but-not-fully-processed operations in a dedicated AtomicInteger:

• incremented in submitOperation() before the operation is enqueued, and rolled back if the operation is rejected (queue full, shutdown, interrupt);
• decremented via the new package-private operationDone(), called from a try/finally around the processing block in TraditionalWorkerThread (covers the normal path, the transaction paths and exceptions);
• decremented for operations cancelled while the queue is being resized (interrupted re-submission path);
• reset in initializeWorkQueue().

isIdle() now simply checks that this counter is zero, covering queued operations, in-flight operations and the hand-off window in between.

The four waitUntilIdle* tests are moved out of the slow group into the default build (~16s for the class).

Verification

TraditionalWorkQueueTestCase run 5 times in a row under the default CI group filter: 6/6 green each time.