What if pass downtime broker nodes to serviceUrl parameter when building PulsarClient

[AuroraTwinkle]

I have opened an issue, please refer to the link:#22934

I am a beginner in Pulsar, and I found a problem. When I create a PulsarClient, if I pass a serviceUrl that contains multiple downtime broker nodes and a normal broker node, then creating a Consumer or Producer will most likely fail and throw an exception similar to the following:

PulsarClient newPulsarClient = PulsarClient.builder()
             .serviceUrl("pulsar://192.168.139.28:6650,192.168.139.28:6650,192.168.139.2:6650,192.168.139.23:6650,192.168.58.4:6650")
             .build();

Exception in thread "main" org.apache.pulsar.client.api.PulsarClientException: java.util.concurrent.ExecutionException: org.apache.pulsar.client.api.PulsarClientException: Failed to subscribe persistent://benchmark/ns-tPALz4g/test-0000000-jbw3YDo with 4 partitions
Connection already closed
	at org.apache.pulsar.client.api.PulsarClientException.unwrap(PulsarClientException.java:1111)
	at org.apache.pulsar.client.impl.ConsumerBuilderImpl.subscribe(ConsumerBuilderImpl.java:103)

By studying the source code, I found that when PulsarClient establishes a connection with the broker, it selects a node in the serviceUrl for connection in a round-robin order.

    @Override
    public InetSocketAddress resolveHost() {
        List<InetSocketAddress> list = addressList;
        checkState(
            list != null, "No service url is provided yet");
        checkState(
            !list.isEmpty(), "No hosts found for service url : " + serviceUrl);
        if (list.size() == 1) {
            return list.get(0);
        } else {
            int originalIndex = CURRENT_INDEX_UPDATER.getAndUpdate(this, last -> (last + 1) % list.size());
            return list.get((originalIndex + 1) % list.size());
        }
    }

In my opinion, as long as there is a healthy broker node in serviceUrl, PulsarClient should be able to create Consumer or Producer normally. So I think there are some areas for improvement here. In the implementation of the ServiceNameResolver interface, we can perform periodic node detection and corresponding removal or recovery operations to avoid selecting unhealthy nodes when resolvingHost, thereby improving the robustness of the program. If you think this is an improvement, I will push a PR to solve this problem. Looking forward to your suggestions, thanks!