PYTHON-5505 added the overload retry loop for all operations but neglects to implement the deprioritized server selection logic. The spec calls on us to deprioritize previous servers for overload retry attempts.

The most straightforward way to do this may be to replace the @_retry_overload decorator with calls to client._retryable_read and client._retryable_write with the "retryable=False" flag to indicate that these operation only attempt retries on overload errors, not on retryable read/writes errors. This is what we do for update_many and delete_many which also go through client._retryable_write.