CassandraSinkCluster: Document desired error handling

[rukai]

This documents what shotover's error handling for CassandraSinkCluster should look like.
Currently we dont handle in the way described here, but I wanted us to agree on an error handling approach that seemed reasonable and that we can work towards. We can of course change things later if we find we need to though.

We could also return an overloaded error when a connection error occurs before sending the message as that would simplify shotover by letting us push retries onto the client.