KafkaProducer potential hang on close() when task.drop.producer.errors==true

The issue is caused by 2 bad behaviors:
1. KafkaProducer.close() does an unbounded thread.join() even when a force close (timeout == 0 ms) is specified.
2. KafkaSystemProducer.handleSendException() has a synchronized block to recreate the producer when task.drop.producer.errors==true. When the IO thread is waiting on the synchronized block, it prevents the KafkaProducer from joining the IO thread.
Fixing either one will fix the issue, but we will close this ticket as soon as the Samza side is tested and committed.