Currently retry_writes option is considered a non-CRUD option in cluster parlance, which means when it's given to Client#with a new cluster in created. As far as I can see there is no reason to create a new cluster when retrying or not retrying writes, therefore this option should be added to the list of CRUD options.

Additionally, CRUD options should not be passed to cluster because a cluster is reused for multiple clients with different CRUD options.

Making these changes requires adjusting which clients some tests use due to ruby-1460.