Fix intermittent failure (final I swear)

[rukai]

I had originally thought that failures were a rare edge case that we should handle individually.
However its become clear that approach wont work.

To fix this issue once and for all I have replaced the illadvised SchemaAwaiter abstraction with logic in our CassandraConnection::execute method that will check for any schema altering commands and call await_schema_agreement() to block until the schema altering command is truly complete.

The rest of the PR should be just asyncification noise

[rukai]

Still getting failures on this?
I wonder if a better fix would be to add a retry mechanism to CassandraConnection::execute?

Or maybe the problem is just that I am reusing the shotover address rather than connecting directly to the cluster.
Maybe a setter API would let us optionally set the schema waiter without making it easy to get the shotover/direct addresses mixed up.

Edit: Ended up solving this with the setter API.