S3A client retries on SSL Auth exceptions triggered by "." bucket names

If you have a "." in bucket names (it's allowed!) then virtual host HTTPS connections fail with a java.net.ssl exception. Except we retry and the inner cause is wrapped by generic "client exceptions"

I'm not going to try and be clever about fixing this, but we should

• make sure that the inner exception is raised up
• avoid retries
• document it in the troubleshooting page.
• if there is a well known public "." bucket (cloudera has some) we can test

I get a vague suspicion the AWS SDK is retrying too. Not much we can do there.