PersistencePollingHandler should not flood the logs with redundant warnings

When a connection is closed, the persistence polling handler logs one warning for each vbucket. Each warning includes a stack trace, so for 1024 vbuckets we're talking about a lot of log lines.

This causes problems in the Elasticsearch connector integration tests, where the server sometimes takes a while to warm up before it will admit to owning vbuckets. The resulting log output is many megabytes and causes the test runner to run out of memory.

The only log message we really care about is the first one, because it describes why the poller closed the connection. The remaining failures are triggered by the connection closure, and don't contain any useful information.