In version 2.3.6 and earlier, an attempt to create a Bucket or a
ClusterInfo would fail fast with a ConnectException in the cause of
the Couchbase server was done. In 2.3.7 this remains true for a
Bucket but is no longer the case for ClusterInfo. The latter now
retries for 75 seconds by default and when it eventually fails a
ConnectException is no longer part of the cause chain.

This commit makes the auto-configured ClusterInfo depend on the
auto-configured Bucket. This means that the Bucket is always created
before the ClusterInfo, thereby ensuring that things fail gracefully
with useful diagnostics when the server is unavailable.

Closes gh-8092