Fix flaky SharedTsBlockQueue concurrency test

[Caideyipi]

Description

SharedTsBlockQueueTest.concurrencyTest used a fixed JUnit timeout while polling counters, printing progress, and recursively scheduling listener tasks. Under CI load, the test could time out without surfacing worker failures, and its executor was never shut down.

This change makes the test deterministic by:

• waiting for producer and consumer tasks through Future.get
• propagating worker failures to the test thread
• using the real memory manager and pool instead of unnecessary spies
• asserting the final queue and memory state
• always shutting down the executor

Testing

• mvn -o -nsu -Ddevelocity.off=true spotless:check -pl iotdb-core/datanode
• compiled and ran SharedTsBlockQueueTest directly: 2 tests passed
• repeated the complete test class 20 times: 20/20 passed

A full reactor test attempt was blocked before reaching DataNode by an existing generated Thrift compilation error: TaskType.java could not resolve javax.annotation.

[JackieTien97]

One non-blocking suggestion to improve failure diagnostics.

[JackieTien97]

Could we wait for whichever worker completes first here? If the receiver fails before the sender finishes, it stops freeing queue memory; the sender can then block once the pool fills, so this Future.get call reports a 30-second TimeoutException and receiver.get is never reached. That hides the receiver's actual exception, even though this change aims to propagate worker failures. Using an ExecutorCompletionService with a shared deadline (and cancelling the sibling task on failure) would make this fail fast and preserve the root cause.