fix: Modify existing mechanism of logging errors in gax-batching to avoid looped calls

[minherz]

Existing error reporting mechanism in gax-batching uses JUL (java.util.logger) to log the errors. This mechanism can lead to the looped calls between the Java client logging library and gax in a case users configure the com.google.cloud.logging.LoggingHandler to serve as one of the JUL handlers. In such a case a call to write a log via JUL will land into gax-batching code (via java-logging). If an error arise, the gax-batching will call JUL to write an error log which, in turn, will return the flow back to the same gax-batching code.
To avoid such situation before the first call the current JUL handlers should be enumerated to ensure they do not include com.google.cloud.logging.LoggingHandler. If the latter occurs in the list, the error message should be logged to STDERR.

More info about batching modification can be found in the internal document.

[chanseokoh]

After talking to @vam-google, I learned that gax-batching is not the only place where gax-java does logging. Places I found:

• BasicRetryingFuture
• GrpcClientCalls: here and here
• RefreshingManagedChannel
• Watchdog
• BatcherImpl