Cluster level ops can error in weird ways if used before connections initialized

When performing cluster level ops without using WaitUntilReady then an op can be made before the SDK connections are initialized. When this happens it doesn't matter what version the cluster is we respond with an error because as far as gocbcore is concerned the agent doesn't support GCCCP.

We should improve this by moving all the GCCCP detection logic out of gocb. It should be handled within gocbcore where it can spin until a valid config is seen or the timeout is reached. Upon valid config seen gocbcore can then proceed or respond with the unsupported error.