Ensure proper retry and backoff for newly created monitor topics

[gitlw]

As shown in the conversations https://linkedin-randd.slack.com/archives/C04FMP0HB17/p1671222219329569,
if a new monitoring topic is just created in a cluster, the AdminClient.describeTopic API could result in UnknownTopicOrPartitionExceptions, which causes the whole process to crash. Below are the places that can trigger the exception (and there maybe more call sites)

kafka-monitor/src/main/java/com/linkedin/xinfra/monitor/services/MultiClusterTopicManagementService.java

Line 455 in 7f99c09

_adminClient.describeTopics(Collections.singleton(_topic)).all().get().get(_topic).partitions();

kafka-monitor/src/main/java/com/linkedin/xinfra/monitor/services/MultiClusterTopicManagementService.java

Line 338 in 7f99c09

List<TopicPartitionInfo> partitions = topicDescriptions.get(_requestTimeout.toMillis(), TimeUnit.MILLISECONDS).partitions();

We need to make sure that the logic calling the describeTopic API has appropriate retries and backoffs in case it's a topic that's just created.

[github-actions]

This is your first issue in the repository. Thank you for raising this issue.' first issue