[fix][client] Producer uses pulsar-client-internal thread to execute callback

[nodece]

Fixes #21900

Motivation

In #21900, I guess the user uses a sync call in the pulsar-client-io callback thread, so throw the timeout exception.

The pulsar-client-io thread is used to communicate between the broker and the client, when pulsar-client-io is stuck, the timeout exception will be thrown.

If we switch from the pulsar-client-io to the pulsar-client-internal thread to execute the callback, #21900 will be fixed.

In the consumer, there's a similar approach：

pulsar/pulsar-client/src/main/java/org/apache/pulsar/client/impl/ConsumerImpl.java

Lines 471 to 487 in 252509e

	protected CompletableFuture<Message<T>> internalReceiveAsync() {
	CompletableFutureCancellationHandler cancellationHandler = new CompletableFutureCancellationHandler();
	CompletableFuture<Message<T>> result = cancellationHandler.createFuture();
	internalPinnedExecutor.execute(() -> {
	Message<T> message = incomingMessages.poll();
	if (message == null) {
	expectMoreIncomingMessages();
	pendingReceives.add(result);
	cancellationHandler.setCancelAction(() -> pendingReceives.remove(result));
	} else {
	messageProcessed(message);
	result.complete(beforeConsume(message));
	}
	});
	return result;
	}

Another approach: Please use the same call(sync or async) in your project.

Modifications

• Producers always use the pulsar-client-internal thread to execute the callback.

Verifying this change

ProducerCallbackThreadTest has been added.

Documentation

• doc
• doc-required
• doc-not-needed
• doc-complete

[nodece]

Some tests fail when multiple threads are used to execute callbacks.

[nodece]

/pulsarbot rerun-failure-checks