KAFKA-21051: testShareConsumerAfterCoordinatorMovement can halt the test JVM

[Russole]

This PR fixes a race in testShareConsumerAfterCoordinatorMovement
where an empty poll could stop the consumer before all produced records
were consumed. It separates the producer stop request from actual
completion, waits for all produced keys, and ensures asynchronous tasks
finish before cleanup.

The issue was reproduced by temporarily scheduling coordinator movement
at 8 seconds. The original test failed frequently, while the fixed test
passed all repeated runs under the same timing.

Test

./gradlew clients:clients-integration-tests:test \
  --tests
org.apache.kafka.clients.consumer.ShareConsumerTest.testShareConsumerAfterCoordinatorMovement
\
  -Pkafka.cluster.test.repeat=5 \
  --rerun-tasks